xref: /openbmc/linux/arch/ia64/kernel/smpboot.c (revision adb57164)
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/pgalloc.h>
53 #include <asm/pgtable.h>
54 #include <asm/processor.h>
55 #include <asm/ptrace.h>
56 #include <asm/sal.h>
57 #include <asm/tlbflush.h>
58 #include <asm/unistd.h>
59 
60 #define SMP_DEBUG 0
61 
62 #if SMP_DEBUG
63 #define Dprintk(x...)  printk(x)
64 #else
65 #define Dprintk(x...)
66 #endif
67 
68 #ifdef CONFIG_HOTPLUG_CPU
69 #ifdef CONFIG_PERMIT_BSP_REMOVE
70 #define bsp_remove_ok	1
71 #else
72 #define bsp_remove_ok	0
73 #endif
74 
75 /*
76  * Global array allocated for NR_CPUS at boot time
77  */
78 struct sal_to_os_boot sal_boot_rendez_state[NR_CPUS];
79 
80 /*
81  * start_ap in head.S uses this to store current booting cpu
82  * info.
83  */
84 struct sal_to_os_boot *sal_state_for_booting_cpu = &sal_boot_rendez_state[0];
85 
86 #define set_brendez_area(x) (sal_state_for_booting_cpu = &sal_boot_rendez_state[(x)]);
87 
88 #else
89 #define set_brendez_area(x)
90 #endif
91 
92 
93 /*
94  * ITC synchronization related stuff:
95  */
96 #define MASTER	(0)
97 #define SLAVE	(SMP_CACHE_BYTES/8)
98 
99 #define NUM_ROUNDS	64	/* magic value */
100 #define NUM_ITERS	5	/* likewise */
101 
102 static DEFINE_SPINLOCK(itc_sync_lock);
103 static volatile unsigned long go[SLAVE + 1];
104 
105 #define DEBUG_ITC_SYNC	0
106 
107 extern void start_ap (void);
108 extern unsigned long ia64_iobase;
109 
110 struct task_struct *task_for_booting_cpu;
111 
112 /*
113  * State for each CPU
114  */
115 DEFINE_PER_CPU(int, cpu_state);
116 
117 cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned;
118 EXPORT_SYMBOL(cpu_core_map);
119 DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map);
120 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
121 
122 int smp_num_siblings = 1;
123 
124 /* which logical CPU number maps to which CPU (physical APIC ID) */
125 volatile int ia64_cpu_to_sapicid[NR_CPUS];
126 EXPORT_SYMBOL(ia64_cpu_to_sapicid);
127 
128 static cpumask_t cpu_callin_map;
129 
130 struct smp_boot_data smp_boot_data __initdata;
131 
132 unsigned long ap_wakeup_vector = -1; /* External Int use to wakeup APs */
133 
134 char __initdata no_int_routing;
135 
136 unsigned char smp_int_redirect; /* are INT and IPI redirectable by the chipset? */
137 
138 #ifdef CONFIG_FORCE_CPEI_RETARGET
139 #define CPEI_OVERRIDE_DEFAULT	(1)
140 #else
141 #define CPEI_OVERRIDE_DEFAULT	(0)
142 #endif
143 
144 unsigned int force_cpei_retarget = CPEI_OVERRIDE_DEFAULT;
145 
146 static int __init
147 cmdl_force_cpei(char *str)
148 {
149 	int value=0;
150 
151 	get_option (&str, &value);
152 	force_cpei_retarget = value;
153 
154 	return 1;
155 }
156 
157 __setup("force_cpei=", cmdl_force_cpei);
158 
159 static int __init
160 nointroute (char *str)
161 {
162 	no_int_routing = 1;
163 	printk ("no_int_routing on\n");
164 	return 1;
165 }
166 
167 __setup("nointroute", nointroute);
168 
169 static void fix_b0_for_bsp(void)
170 {
171 #ifdef CONFIG_HOTPLUG_CPU
172 	int cpuid;
173 	static int fix_bsp_b0 = 1;
174 
175 	cpuid = smp_processor_id();
176 
177 	/*
178 	 * Cache the b0 value on the first AP that comes up
179 	 */
180 	if (!(fix_bsp_b0 && cpuid))
181 		return;
182 
183 	sal_boot_rendez_state[0].br[0] = sal_boot_rendez_state[cpuid].br[0];
184 	printk ("Fixed BSP b0 value from CPU %d\n", cpuid);
185 
186 	fix_bsp_b0 = 0;
187 #endif
188 }
189 
190 void
191 sync_master (void *arg)
192 {
193 	unsigned long flags, i;
194 
195 	go[MASTER] = 0;
196 
197 	local_irq_save(flags);
198 	{
199 		for (i = 0; i < NUM_ROUNDS*NUM_ITERS; ++i) {
200 			while (!go[MASTER])
201 				cpu_relax();
202 			go[MASTER] = 0;
203 			go[SLAVE] = ia64_get_itc();
204 		}
205 	}
206 	local_irq_restore(flags);
207 }
208 
209 /*
210  * Return the number of cycles by which our itc differs from the itc on the master
211  * (time-keeper) CPU.  A positive number indicates our itc is ahead of the master,
212  * negative that it is behind.
213  */
214 static inline long
215 get_delta (long *rt, long *master)
216 {
217 	unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
218 	unsigned long tcenter, t0, t1, tm;
219 	long i;
220 
221 	for (i = 0; i < NUM_ITERS; ++i) {
222 		t0 = ia64_get_itc();
223 		go[MASTER] = 1;
224 		while (!(tm = go[SLAVE]))
225 			cpu_relax();
226 		go[SLAVE] = 0;
227 		t1 = ia64_get_itc();
228 
229 		if (t1 - t0 < best_t1 - best_t0)
230 			best_t0 = t0, best_t1 = t1, best_tm = tm;
231 	}
232 
233 	*rt = best_t1 - best_t0;
234 	*master = best_tm - best_t0;
235 
236 	/* average best_t0 and best_t1 without overflow: */
237 	tcenter = (best_t0/2 + best_t1/2);
238 	if (best_t0 % 2 + best_t1 % 2 == 2)
239 		++tcenter;
240 	return tcenter - best_tm;
241 }
242 
243 /*
244  * Synchronize ar.itc of the current (slave) CPU with the ar.itc of the MASTER CPU
245  * (normally the time-keeper CPU).  We use a closed loop to eliminate the possibility of
246  * unaccounted-for errors (such as getting a machine check in the middle of a calibration
247  * step).  The basic idea is for the slave to ask the master what itc value it has and to
248  * read its own itc before and after the master responds.  Each iteration gives us three
249  * timestamps:
250  *
251  *	slave		master
252  *
253  *	t0 ---\
254  *             ---\
255  *		   --->
256  *			tm
257  *		   /---
258  *	       /---
259  *	t1 <---
260  *
261  *
262  * The goal is to adjust the slave's ar.itc such that tm falls exactly half-way between t0
263  * and t1.  If we achieve this, the clocks are synchronized provided the interconnect
264  * between the slave and the master is symmetric.  Even if the interconnect were
265  * asymmetric, we would still know that the synchronization error is smaller than the
266  * roundtrip latency (t0 - t1).
267  *
268  * When the interconnect is quiet and symmetric, this lets us synchronize the itc to
269  * within one or two cycles.  However, we can only *guarantee* that the synchronization is
270  * accurate to within a round-trip time, which is typically in the range of several
271  * hundred cycles (e.g., ~500 cycles).  In practice, this means that the itc's are usually
272  * almost perfectly synchronized, but we shouldn't assume that the accuracy is much better
273  * than half a micro second or so.
274  */
275 void
276 ia64_sync_itc (unsigned int master)
277 {
278 	long i, delta, adj, adjust_latency = 0, done = 0;
279 	unsigned long flags, rt, master_time_stamp, bound;
280 #if DEBUG_ITC_SYNC
281 	struct {
282 		long rt;	/* roundtrip time */
283 		long master;	/* master's timestamp */
284 		long diff;	/* difference between midpoint and master's timestamp */
285 		long lat;	/* estimate of itc adjustment latency */
286 	} t[NUM_ROUNDS];
287 #endif
288 
289 	/*
290 	 * Make sure local timer ticks are disabled while we sync.  If
291 	 * they were enabled, we'd have to worry about nasty issues
292 	 * like setting the ITC ahead of (or a long time before) the
293 	 * next scheduled tick.
294 	 */
295 	BUG_ON((ia64_get_itv() & (1 << 16)) == 0);
296 
297 	go[MASTER] = 1;
298 
299 	if (smp_call_function_single(master, sync_master, NULL, 0) < 0) {
300 		printk(KERN_ERR "sync_itc: failed to get attention of CPU %u!\n", master);
301 		return;
302 	}
303 
304 	while (go[MASTER])
305 		cpu_relax();	/* wait for master to be ready */
306 
307 	spin_lock_irqsave(&itc_sync_lock, flags);
308 	{
309 		for (i = 0; i < NUM_ROUNDS; ++i) {
310 			delta = get_delta(&rt, &master_time_stamp);
311 			if (delta == 0) {
312 				done = 1;	/* let's lock on to this... */
313 				bound = rt;
314 			}
315 
316 			if (!done) {
317 				if (i > 0) {
318 					adjust_latency += -delta;
319 					adj = -delta + adjust_latency/4;
320 				} else
321 					adj = -delta;
322 
323 				ia64_set_itc(ia64_get_itc() + adj);
324 			}
325 #if DEBUG_ITC_SYNC
326 			t[i].rt = rt;
327 			t[i].master = master_time_stamp;
328 			t[i].diff = delta;
329 			t[i].lat = adjust_latency/4;
330 #endif
331 		}
332 	}
333 	spin_unlock_irqrestore(&itc_sync_lock, flags);
334 
335 #if DEBUG_ITC_SYNC
336 	for (i = 0; i < NUM_ROUNDS; ++i)
337 		printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
338 		       t[i].rt, t[i].master, t[i].diff, t[i].lat);
339 #endif
340 
341 	printk(KERN_INFO "CPU %d: synchronized ITC with CPU %u (last diff %ld cycles, "
342 	       "maxerr %lu cycles)\n", smp_processor_id(), master, delta, rt);
343 }
344 
345 /*
346  * Ideally sets up per-cpu profiling hooks.  Doesn't do much now...
347  */
348 static inline void smp_setup_percpu_timer(void)
349 {
350 }
351 
352 static void
353 smp_callin (void)
354 {
355 	int cpuid, phys_id, itc_master;
356 	struct cpuinfo_ia64 *last_cpuinfo, *this_cpuinfo;
357 	extern void ia64_init_itm(void);
358 	extern volatile int time_keeper_id;
359 
360 #ifdef CONFIG_PERFMON
361 	extern void pfm_init_percpu(void);
362 #endif
363 
364 	cpuid = smp_processor_id();
365 	phys_id = hard_smp_processor_id();
366 	itc_master = time_keeper_id;
367 
368 	if (cpu_online(cpuid)) {
369 		printk(KERN_ERR "huh, phys CPU#0x%x, CPU#0x%x already present??\n",
370 		       phys_id, cpuid);
371 		BUG();
372 	}
373 
374 	fix_b0_for_bsp();
375 
376 	/*
377 	 * numa_node_id() works after this.
378 	 */
379 	set_numa_node(cpu_to_node_map[cpuid]);
380 	set_numa_mem(local_memory_node(cpu_to_node_map[cpuid]));
381 
382 	spin_lock(&vector_lock);
383 	/* Setup the per cpu irq handling data structures */
384 	__setup_vector_irq(cpuid);
385 	notify_cpu_starting(cpuid);
386 	set_cpu_online(cpuid, true);
387 	per_cpu(cpu_state, cpuid) = CPU_ONLINE;
388 	spin_unlock(&vector_lock);
389 
390 	smp_setup_percpu_timer();
391 
392 	ia64_mca_cmc_vector_setup();	/* Setup vector on AP */
393 
394 #ifdef CONFIG_PERFMON
395 	pfm_init_percpu();
396 #endif
397 
398 	local_irq_enable();
399 
400 	if (!(sal_platform_features & IA64_SAL_PLATFORM_FEATURE_ITC_DRIFT)) {
401 		/*
402 		 * Synchronize the ITC with the BP.  Need to do this after irqs are
403 		 * enabled because ia64_sync_itc() calls smp_call_function_single(), which
404 		 * calls spin_unlock_bh(), which calls spin_unlock_bh(), which calls
405 		 * local_bh_enable(), which bugs out if irqs are not enabled...
406 		 */
407 		Dprintk("Going to syncup ITC with ITC Master.\n");
408 		ia64_sync_itc(itc_master);
409 	}
410 
411 	/*
412 	 * Get our bogomips.
413 	 */
414 	ia64_init_itm();
415 
416 	/*
417 	 * Delay calibration can be skipped if new processor is identical to the
418 	 * previous processor.
419 	 */
420 	last_cpuinfo = cpu_data(cpuid - 1);
421 	this_cpuinfo = local_cpu_data;
422 	if (last_cpuinfo->itc_freq != this_cpuinfo->itc_freq ||
423 	    last_cpuinfo->proc_freq != this_cpuinfo->proc_freq ||
424 	    last_cpuinfo->features != this_cpuinfo->features ||
425 	    last_cpuinfo->revision != this_cpuinfo->revision ||
426 	    last_cpuinfo->family != this_cpuinfo->family ||
427 	    last_cpuinfo->archrev != this_cpuinfo->archrev ||
428 	    last_cpuinfo->model != this_cpuinfo->model)
429 		calibrate_delay();
430 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
431 
432 	/*
433 	 * Allow the master to continue.
434 	 */
435 	cpumask_set_cpu(cpuid, &cpu_callin_map);
436 	Dprintk("Stack on CPU %d at about %p\n",cpuid, &cpuid);
437 }
438 
439 
440 /*
441  * Activate a secondary processor.  head.S calls this.
442  */
443 int
444 start_secondary (void *unused)
445 {
446 	/* Early console may use I/O ports */
447 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
448 #ifndef CONFIG_PRINTK_TIME
449 	Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id());
450 #endif
451 	efi_map_pal_code();
452 	cpu_init();
453 	preempt_disable();
454 	smp_callin();
455 
456 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
457 	return 0;
458 }
459 
460 static int
461 do_boot_cpu (int sapicid, int cpu, struct task_struct *idle)
462 {
463 	int timeout;
464 
465 	task_for_booting_cpu = idle;
466 	Dprintk("Sending wakeup vector %lu to AP 0x%x/0x%x.\n", ap_wakeup_vector, cpu, sapicid);
467 
468 	set_brendez_area(cpu);
469 	ia64_send_ipi(cpu, ap_wakeup_vector, IA64_IPI_DM_INT, 0);
470 
471 	/*
472 	 * Wait 10s total for the AP to start
473 	 */
474 	Dprintk("Waiting on callin_map ...");
475 	for (timeout = 0; timeout < 100000; timeout++) {
476 		if (cpumask_test_cpu(cpu, &cpu_callin_map))
477 			break;  /* It has booted */
478 		barrier(); /* Make sure we re-read cpu_callin_map */
479 		udelay(100);
480 	}
481 	Dprintk("\n");
482 
483 	if (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
484 		printk(KERN_ERR "Processor 0x%x/0x%x is stuck.\n", cpu, sapicid);
485 		ia64_cpu_to_sapicid[cpu] = -1;
486 		set_cpu_online(cpu, false);  /* was set in smp_callin() */
487 		return -EINVAL;
488 	}
489 	return 0;
490 }
491 
492 static int __init
493 decay (char *str)
494 {
495 	int ticks;
496 	get_option (&str, &ticks);
497 	return 1;
498 }
499 
500 __setup("decay=", decay);
501 
502 /*
503  * Initialize the logical CPU number to SAPICID mapping
504  */
505 void __init
506 smp_build_cpu_map (void)
507 {
508 	int sapicid, cpu, i;
509 	int boot_cpu_id = hard_smp_processor_id();
510 
511 	for (cpu = 0; cpu < NR_CPUS; cpu++) {
512 		ia64_cpu_to_sapicid[cpu] = -1;
513 	}
514 
515 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
516 	init_cpu_present(cpumask_of(0));
517 	set_cpu_possible(0, true);
518 	for (cpu = 1, i = 0; i < smp_boot_data.cpu_count; i++) {
519 		sapicid = smp_boot_data.cpu_phys_id[i];
520 		if (sapicid == boot_cpu_id)
521 			continue;
522 		set_cpu_present(cpu, true);
523 		set_cpu_possible(cpu, true);
524 		ia64_cpu_to_sapicid[cpu] = sapicid;
525 		cpu++;
526 	}
527 }
528 
529 /*
530  * Cycle through the APs sending Wakeup IPIs to boot each.
531  */
532 void __init
533 smp_prepare_cpus (unsigned int max_cpus)
534 {
535 	int boot_cpu_id = hard_smp_processor_id();
536 
537 	/*
538 	 * Initialize the per-CPU profiling counter/multiplier
539 	 */
540 
541 	smp_setup_percpu_timer();
542 
543 	cpumask_set_cpu(0, &cpu_callin_map);
544 
545 	local_cpu_data->loops_per_jiffy = loops_per_jiffy;
546 	ia64_cpu_to_sapicid[0] = boot_cpu_id;
547 
548 	printk(KERN_INFO "Boot processor id 0x%x/0x%x\n", 0, boot_cpu_id);
549 
550 	current_thread_info()->cpu = 0;
551 
552 	/*
553 	 * If SMP should be disabled, then really disable it!
554 	 */
555 	if (!max_cpus) {
556 		printk(KERN_INFO "SMP mode deactivated.\n");
557 		init_cpu_online(cpumask_of(0));
558 		init_cpu_present(cpumask_of(0));
559 		init_cpu_possible(cpumask_of(0));
560 		return;
561 	}
562 }
563 
564 void smp_prepare_boot_cpu(void)
565 {
566 	set_cpu_online(smp_processor_id(), true);
567 	cpumask_set_cpu(smp_processor_id(), &cpu_callin_map);
568 	set_numa_node(cpu_to_node_map[smp_processor_id()]);
569 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
570 }
571 
572 #ifdef CONFIG_HOTPLUG_CPU
573 static inline void
574 clear_cpu_sibling_map(int cpu)
575 {
576 	int i;
577 
578 	for_each_cpu(i, &per_cpu(cpu_sibling_map, cpu))
579 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, i));
580 	for_each_cpu(i, &cpu_core_map[cpu])
581 		cpumask_clear_cpu(cpu, &cpu_core_map[i]);
582 
583 	per_cpu(cpu_sibling_map, cpu) = cpu_core_map[cpu] = CPU_MASK_NONE;
584 }
585 
586 static void
587 remove_siblinginfo(int cpu)
588 {
589 	int last = 0;
590 
591 	if (cpu_data(cpu)->threads_per_core == 1 &&
592 	    cpu_data(cpu)->cores_per_socket == 1) {
593 		cpumask_clear_cpu(cpu, &cpu_core_map[cpu]);
594 		cpumask_clear_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
595 		return;
596 	}
597 
598 	last = (cpumask_weight(&cpu_core_map[cpu]) == 1 ? 1 : 0);
599 
600 	/* remove it from all sibling map's */
601 	clear_cpu_sibling_map(cpu);
602 }
603 
604 extern void fixup_irqs(void);
605 
606 int migrate_platform_irqs(unsigned int cpu)
607 {
608 	int new_cpei_cpu;
609 	struct irq_data *data = NULL;
610 	const struct cpumask *mask;
611 	int 		retval = 0;
612 
613 	/*
614 	 * dont permit CPEI target to removed.
615 	 */
616 	if (cpe_vector > 0 && is_cpu_cpei_target(cpu)) {
617 		printk ("CPU (%d) is CPEI Target\n", cpu);
618 		if (can_cpei_retarget()) {
619 			/*
620 			 * Now re-target the CPEI to a different processor
621 			 */
622 			new_cpei_cpu = cpumask_any(cpu_online_mask);
623 			mask = cpumask_of(new_cpei_cpu);
624 			set_cpei_target_cpu(new_cpei_cpu);
625 			data = irq_get_irq_data(ia64_cpe_irq);
626 			/*
627 			 * Switch for now, immediately, we need to do fake intr
628 			 * as other interrupts, but need to study CPEI behaviour with
629 			 * polling before making changes.
630 			 */
631 			if (data && data->chip) {
632 				data->chip->irq_disable(data);
633 				data->chip->irq_set_affinity(data, mask, false);
634 				data->chip->irq_enable(data);
635 				printk ("Re-targeting CPEI to cpu %d\n", new_cpei_cpu);
636 			}
637 		}
638 		if (!data) {
639 			printk ("Unable to retarget CPEI, offline cpu [%d] failed\n", cpu);
640 			retval = -EBUSY;
641 		}
642 	}
643 	return retval;
644 }
645 
646 /* must be called with cpucontrol mutex held */
647 int __cpu_disable(void)
648 {
649 	int cpu = smp_processor_id();
650 
651 	/*
652 	 * dont permit boot processor for now
653 	 */
654 	if (cpu == 0 && !bsp_remove_ok) {
655 		printk ("Your platform does not support removal of BSP\n");
656 		return (-EBUSY);
657 	}
658 
659 	set_cpu_online(cpu, false);
660 
661 	if (migrate_platform_irqs(cpu)) {
662 		set_cpu_online(cpu, true);
663 		return -EBUSY;
664 	}
665 
666 	remove_siblinginfo(cpu);
667 	fixup_irqs();
668 	local_flush_tlb_all();
669 	cpumask_clear_cpu(cpu, &cpu_callin_map);
670 	return 0;
671 }
672 
673 void __cpu_die(unsigned int cpu)
674 {
675 	unsigned int i;
676 
677 	for (i = 0; i < 100; i++) {
678 		/* They ack this in play_dead by setting CPU_DEAD */
679 		if (per_cpu(cpu_state, cpu) == CPU_DEAD)
680 		{
681 			printk ("CPU %d is now offline\n", cpu);
682 			return;
683 		}
684 		msleep(100);
685 	}
686  	printk(KERN_ERR "CPU %u didn't die...\n", cpu);
687 }
688 #endif /* CONFIG_HOTPLUG_CPU */
689 
690 void
691 smp_cpus_done (unsigned int dummy)
692 {
693 	int cpu;
694 	unsigned long bogosum = 0;
695 
696 	/*
697 	 * Allow the user to impress friends.
698 	 */
699 
700 	for_each_online_cpu(cpu) {
701 		bogosum += cpu_data(cpu)->loops_per_jiffy;
702 	}
703 
704 	printk(KERN_INFO "Total of %d processors activated (%lu.%02lu BogoMIPS).\n",
705 	       (int)num_online_cpus(), bogosum/(500000/HZ), (bogosum/(5000/HZ))%100);
706 }
707 
708 static inline void set_cpu_sibling_map(int cpu)
709 {
710 	int i;
711 
712 	for_each_online_cpu(i) {
713 		if ((cpu_data(cpu)->socket_id == cpu_data(i)->socket_id)) {
714 			cpumask_set_cpu(i, &cpu_core_map[cpu]);
715 			cpumask_set_cpu(cpu, &cpu_core_map[i]);
716 			if (cpu_data(cpu)->core_id == cpu_data(i)->core_id) {
717 				cpumask_set_cpu(i,
718 						&per_cpu(cpu_sibling_map, cpu));
719 				cpumask_set_cpu(cpu,
720 						&per_cpu(cpu_sibling_map, i));
721 			}
722 		}
723 	}
724 }
725 
726 int
727 __cpu_up(unsigned int cpu, struct task_struct *tidle)
728 {
729 	int ret;
730 	int sapicid;
731 
732 	sapicid = ia64_cpu_to_sapicid[cpu];
733 	if (sapicid == -1)
734 		return -EINVAL;
735 
736 	/*
737 	 * Already booted cpu? not valid anymore since we dont
738 	 * do idle loop tightspin anymore.
739 	 */
740 	if (cpumask_test_cpu(cpu, &cpu_callin_map))
741 		return -EINVAL;
742 
743 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
744 	/* Processor goes to start_secondary(), sets online flag */
745 	ret = do_boot_cpu(sapicid, cpu, tidle);
746 	if (ret < 0)
747 		return ret;
748 
749 	if (cpu_data(cpu)->threads_per_core == 1 &&
750 	    cpu_data(cpu)->cores_per_socket == 1) {
751 		cpumask_set_cpu(cpu, &per_cpu(cpu_sibling_map, cpu));
752 		cpumask_set_cpu(cpu, &cpu_core_map[cpu]);
753 		return 0;
754 	}
755 
756 	set_cpu_sibling_map(cpu);
757 
758 	return 0;
759 }
760 
761 /*
762  * Assume that CPUs have been discovered by some platform-dependent interface.  For
763  * SoftSDV/Lion, that would be ACPI.
764  *
765  * Setup of the IPI irq handler is done in irq.c:init_IRQ_SMP().
766  */
767 void __init
768 init_smp_config(void)
769 {
770 	struct fptr {
771 		unsigned long fp;
772 		unsigned long gp;
773 	} *ap_startup;
774 	long sal_ret;
775 
776 	/* Tell SAL where to drop the APs.  */
777 	ap_startup = (struct fptr *) start_ap;
778 	sal_ret = ia64_sal_set_vectors(SAL_VECTOR_OS_BOOT_RENDEZ,
779 				       ia64_tpa(ap_startup->fp), ia64_tpa(ap_startup->gp), 0, 0, 0, 0);
780 	if (sal_ret < 0)
781 		printk(KERN_ERR "SMP: Can't set SAL AP Boot Rendezvous: %s\n",
782 		       ia64_sal_strerror(sal_ret));
783 }
784 
785 /*
786  * identify_siblings(cpu) gets called from identify_cpu. This populates the
787  * information related to logical execution units in per_cpu_data structure.
788  */
789 void identify_siblings(struct cpuinfo_ia64 *c)
790 {
791 	long status;
792 	u16 pltid;
793 	pal_logical_to_physical_t info;
794 
795 	status = ia64_pal_logical_to_phys(-1, &info);
796 	if (status != PAL_STATUS_SUCCESS) {
797 		if (status != PAL_STATUS_UNIMPLEMENTED) {
798 			printk(KERN_ERR
799 				"ia64_pal_logical_to_phys failed with %ld\n",
800 				status);
801 			return;
802 		}
803 
804 		info.overview_ppid = 0;
805 		info.overview_cpp  = 1;
806 		info.overview_tpc  = 1;
807 	}
808 
809 	status = ia64_sal_physical_id_info(&pltid);
810 	if (status != PAL_STATUS_SUCCESS) {
811 		if (status != PAL_STATUS_UNIMPLEMENTED)
812 			printk(KERN_ERR
813 				"ia64_sal_pltid failed with %ld\n",
814 				status);
815 		return;
816 	}
817 
818 	c->socket_id =  (pltid << 8) | info.overview_ppid;
819 
820 	if (info.overview_cpp == 1 && info.overview_tpc == 1)
821 		return;
822 
823 	c->cores_per_socket = info.overview_cpp;
824 	c->threads_per_core = info.overview_tpc;
825 	c->num_log = info.overview_num_log;
826 
827 	c->core_id = info.log1_cid;
828 	c->thread_id = info.log1_tid;
829 }
830 
831 /*
832  * returns non zero, if multi-threading is enabled
833  * on at least one physical package. Due to hotplug cpu
834  * and (maxcpus=), all threads may not necessarily be enabled
835  * even though the processor supports multi-threading.
836  */
837 int is_multithreading_enabled(void)
838 {
839 	int i, j;
840 
841 	for_each_present_cpu(i) {
842 		for_each_present_cpu(j) {
843 			if (j == i)
844 				continue;
845 			if ((cpu_data(j)->socket_id == cpu_data(i)->socket_id)) {
846 				if (cpu_data(j)->core_id == cpu_data(i)->core_id)
847 					return 1;
848 			}
849 		}
850 	}
851 	return 0;
852 }
853 EXPORT_SYMBOL_GPL(is_multithreading_enabled);
854