1/*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version 2
5 * of the License, or (at your option) any later version.
6 *
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10 * GNU General Public License for more details.
11 *
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
15 *
16 * Copyright (C) 2000, 2001 Kanoj Sarcar
17 * Copyright (C) 2000, 2001 Ralf Baechle
18 * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
19 * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
20 */
21#include <linux/cache.h>
22#include <linux/delay.h>
23#include <linux/init.h>
24#include <linux/interrupt.h>
25#include <linux/smp.h>
26#include <linux/spinlock.h>
27#include <linux/threads.h>
28#include <linux/module.h>
29#include <linux/time.h>
30#include <linux/timex.h>
31#include <linux/sched.h>
32#include <linux/cpumask.h>
33#include <linux/cpu.h>
34#include <linux/err.h>
35#include <linux/ftrace.h>
36
37#include <linux/atomic.h>
38#include <asm/cpu.h>
39#include <asm/processor.h>
40#include <asm/idle.h>
41#include <asm/r4k-timer.h>
42#include <asm/mmu_context.h>
43#include <asm/time.h>
44#include <asm/setup.h>
45
46cpumask_t cpu_callin_map;		/* Bitmask of started secondaries */
47
48int __cpu_number_map[NR_CPUS];		/* Map physical to logical */
49EXPORT_SYMBOL(__cpu_number_map);
50
51int __cpu_logical_map[NR_CPUS];		/* Map logical to physical */
52EXPORT_SYMBOL(__cpu_logical_map);
53
54/* Number of TCs (or siblings in Intel speak) per CPU core */
55int smp_num_siblings = 1;
56EXPORT_SYMBOL(smp_num_siblings);
57
58/* representing the TCs (or siblings in Intel speak) of each logical CPU */
59cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
60EXPORT_SYMBOL(cpu_sibling_map);
61
62/* representing the core map of multi-core chips of each logical CPU */
63cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
64EXPORT_SYMBOL(cpu_core_map);
65
66/*
67 * A logcal cpu mask containing only one VPE per core to
68 * reduce the number of IPIs on large MT systems.
69 */
70cpumask_t cpu_foreign_map __read_mostly;
71EXPORT_SYMBOL(cpu_foreign_map);
72
73/* representing cpus for which sibling maps can be computed */
74static cpumask_t cpu_sibling_setup_map;
75
76/* representing cpus for which core maps can be computed */
77static cpumask_t cpu_core_setup_map;
78
79cpumask_t cpu_coherent_mask;
80
81static inline void set_cpu_sibling_map(int cpu)
82{
83	int i;
84
85	cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
86
87	if (smp_num_siblings > 1) {
88		for_each_cpu(i, &cpu_sibling_setup_map) {
89			if (cpu_data[cpu].package == cpu_data[i].package &&
90				    cpu_data[cpu].core == cpu_data[i].core) {
91				cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
92				cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
93			}
94		}
95	} else
96		cpumask_set_cpu(cpu, &cpu_sibling_map[cpu]);
97}
98
99static inline void set_cpu_core_map(int cpu)
100{
101	int i;
102
103	cpumask_set_cpu(cpu, &cpu_core_setup_map);
104
105	for_each_cpu(i, &cpu_core_setup_map) {
106		if (cpu_data[cpu].package == cpu_data[i].package) {
107			cpumask_set_cpu(i, &cpu_core_map[cpu]);
108			cpumask_set_cpu(cpu, &cpu_core_map[i]);
109		}
110	}
111}
112
113/*
114 * Calculate a new cpu_foreign_map mask whenever a
115 * new cpu appears or disappears.
116 */
117static inline void calculate_cpu_foreign_map(void)
118{
119	int i, k, core_present;
120	cpumask_t temp_foreign_map;
121
122	/* Re-calculate the mask */
123	cpumask_clear(&temp_foreign_map);
124	for_each_online_cpu(i) {
125		core_present = 0;
126		for_each_cpu(k, &temp_foreign_map)
127			if (cpu_data[i].package == cpu_data[k].package &&
128			    cpu_data[i].core == cpu_data[k].core)
129				core_present = 1;
130		if (!core_present)
131			cpumask_set_cpu(i, &temp_foreign_map);
132	}
133
134	cpumask_copy(&cpu_foreign_map, &temp_foreign_map);
135}
136
137struct plat_smp_ops *mp_ops;
138EXPORT_SYMBOL(mp_ops);
139
140void register_smp_ops(struct plat_smp_ops *ops)
141{
142	if (mp_ops)
143		printk(KERN_WARNING "Overriding previously set SMP ops\n");
144
145	mp_ops = ops;
146}
147
148/*
149 * First C code run on the secondary CPUs after being started up by
150 * the master.
151 */
152asmlinkage void start_secondary(void)
153{
154	unsigned int cpu;
155
156	cpu_probe();
157	per_cpu_trap_init(false);
158	mips_clockevent_init();
159	mp_ops->init_secondary();
160	cpu_report();
161
162	/*
163	 * XXX parity protection should be folded in here when it's converted
164	 * to an option instead of something based on .cputype
165	 */
166
167	calibrate_delay();
168	preempt_disable();
169	cpu = smp_processor_id();
170	cpu_data[cpu].udelay_val = loops_per_jiffy;
171
172	cpumask_set_cpu(cpu, &cpu_coherent_mask);
173	notify_cpu_starting(cpu);
174
175	set_cpu_online(cpu, true);
176
177	set_cpu_sibling_map(cpu);
178	set_cpu_core_map(cpu);
179
180	calculate_cpu_foreign_map();
181
182	cpumask_set_cpu(cpu, &cpu_callin_map);
183
184	synchronise_count_slave(cpu);
185
186	/*
187	 * irq will be enabled in ->smp_finish(), enabling it too early
188	 * is dangerous.
189	 */
190	WARN_ON_ONCE(!irqs_disabled());
191	mp_ops->smp_finish();
192
193	cpu_startup_entry(CPUHP_ONLINE);
194}
195
196/*
197 * Call into both interrupt handlers, as we share the IPI for them
198 */
199void __irq_entry smp_call_function_interrupt(void)
200{
201	irq_enter();
202	generic_smp_call_function_interrupt();
203	irq_exit();
204}
205
206static void stop_this_cpu(void *dummy)
207{
208	/*
209	 * Remove this CPU. Be a bit slow here and
210	 * set the bits for every online CPU so we don't miss
211	 * any IPI whilst taking this VPE down.
212	 */
213
214	cpumask_copy(&cpu_foreign_map, cpu_online_mask);
215
216	/* Make it visible to every other CPU */
217	smp_mb();
218
219	set_cpu_online(smp_processor_id(), false);
220	calculate_cpu_foreign_map();
221	local_irq_disable();
222	while (1);
223}
224
225void smp_send_stop(void)
226{
227	smp_call_function(stop_this_cpu, NULL, 0);
228}
229
230void __init smp_cpus_done(unsigned int max_cpus)
231{
232}
233
234/* called from main before smp_init() */
235void __init smp_prepare_cpus(unsigned int max_cpus)
236{
237	init_new_context(current, &init_mm);
238	current_thread_info()->cpu = 0;
239	mp_ops->prepare_cpus(max_cpus);
240	set_cpu_sibling_map(0);
241	set_cpu_core_map(0);
242	calculate_cpu_foreign_map();
243#ifndef CONFIG_HOTPLUG_CPU
244	init_cpu_present(cpu_possible_mask);
245#endif
246	cpumask_copy(&cpu_coherent_mask, cpu_possible_mask);
247}
248
249/* preload SMP state for boot cpu */
250void smp_prepare_boot_cpu(void)
251{
252	set_cpu_possible(0, true);
253	set_cpu_online(0, true);
254	cpumask_set_cpu(0, &cpu_callin_map);
255}
256
257int __cpu_up(unsigned int cpu, struct task_struct *tidle)
258{
259	mp_ops->boot_secondary(cpu, tidle);
260
261	/*
262	 * Trust is futile.  We should really have timeouts ...
263	 */
264	while (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
265		udelay(100);
266		schedule();
267	}
268
269	synchronise_count_master(cpu);
270	return 0;
271}
272
273/* Not really SMP stuff ... */
274int setup_profiling_timer(unsigned int multiplier)
275{
276	return 0;
277}
278
279static void flush_tlb_all_ipi(void *info)
280{
281	local_flush_tlb_all();
282}
283
284void flush_tlb_all(void)
285{
286	on_each_cpu(flush_tlb_all_ipi, NULL, 1);
287}
288
289static void flush_tlb_mm_ipi(void *mm)
290{
291	local_flush_tlb_mm((struct mm_struct *)mm);
292}
293
294/*
295 * Special Variant of smp_call_function for use by TLB functions:
296 *
297 *  o No return value
298 *  o collapses to normal function call on UP kernels
299 *  o collapses to normal function call on systems with a single shared
300 *    primary cache.
301 */
302static inline void smp_on_other_tlbs(void (*func) (void *info), void *info)
303{
304	smp_call_function(func, info, 1);
305}
306
307static inline void smp_on_each_tlb(void (*func) (void *info), void *info)
308{
309	preempt_disable();
310
311	smp_on_other_tlbs(func, info);
312	func(info);
313
314	preempt_enable();
315}
316
317/*
318 * The following tlb flush calls are invoked when old translations are
319 * being torn down, or pte attributes are changing. For single threaded
320 * address spaces, a new context is obtained on the current cpu, and tlb
321 * context on other cpus are invalidated to force a new context allocation
322 * at switch_mm time, should the mm ever be used on other cpus. For
323 * multithreaded address spaces, intercpu interrupts have to be sent.
324 * Another case where intercpu interrupts are required is when the target
325 * mm might be active on another cpu (eg debuggers doing the flushes on
326 * behalf of debugees, kswapd stealing pages from another process etc).
327 * Kanoj 07/00.
328 */
329
330void flush_tlb_mm(struct mm_struct *mm)
331{
332	preempt_disable();
333
334	if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
335		smp_on_other_tlbs(flush_tlb_mm_ipi, mm);
336	} else {
337		unsigned int cpu;
338
339		for_each_online_cpu(cpu) {
340			if (cpu != smp_processor_id() && cpu_context(cpu, mm))
341				cpu_context(cpu, mm) = 0;
342		}
343	}
344	local_flush_tlb_mm(mm);
345
346	preempt_enable();
347}
348
349struct flush_tlb_data {
350	struct vm_area_struct *vma;
351	unsigned long addr1;
352	unsigned long addr2;
353};
354
355static void flush_tlb_range_ipi(void *info)
356{
357	struct flush_tlb_data *fd = info;
358
359	local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
360}
361
362void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
363{
364	struct mm_struct *mm = vma->vm_mm;
365
366	preempt_disable();
367	if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
368		struct flush_tlb_data fd = {
369			.vma = vma,
370			.addr1 = start,
371			.addr2 = end,
372		};
373
374		smp_on_other_tlbs(flush_tlb_range_ipi, &fd);
375	} else {
376		unsigned int cpu;
377
378		for_each_online_cpu(cpu) {
379			if (cpu != smp_processor_id() && cpu_context(cpu, mm))
380				cpu_context(cpu, mm) = 0;
381		}
382	}
383	local_flush_tlb_range(vma, start, end);
384	preempt_enable();
385}
386
387static void flush_tlb_kernel_range_ipi(void *info)
388{
389	struct flush_tlb_data *fd = info;
390
391	local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
392}
393
394void flush_tlb_kernel_range(unsigned long start, unsigned long end)
395{
396	struct flush_tlb_data fd = {
397		.addr1 = start,
398		.addr2 = end,
399	};
400
401	on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
402}
403
404static void flush_tlb_page_ipi(void *info)
405{
406	struct flush_tlb_data *fd = info;
407
408	local_flush_tlb_page(fd->vma, fd->addr1);
409}
410
411void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
412{
413	preempt_disable();
414	if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
415		struct flush_tlb_data fd = {
416			.vma = vma,
417			.addr1 = page,
418		};
419
420		smp_on_other_tlbs(flush_tlb_page_ipi, &fd);
421	} else {
422		unsigned int cpu;
423
424		for_each_online_cpu(cpu) {
425			if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
426				cpu_context(cpu, vma->vm_mm) = 0;
427		}
428	}
429	local_flush_tlb_page(vma, page);
430	preempt_enable();
431}
432
433static void flush_tlb_one_ipi(void *info)
434{
435	unsigned long vaddr = (unsigned long) info;
436
437	local_flush_tlb_one(vaddr);
438}
439
440void flush_tlb_one(unsigned long vaddr)
441{
442	smp_on_each_tlb(flush_tlb_one_ipi, (void *) vaddr);
443}
444
445EXPORT_SYMBOL(flush_tlb_page);
446EXPORT_SYMBOL(flush_tlb_one);
447
448#if defined(CONFIG_KEXEC)
449void (*dump_ipi_function_ptr)(void *) = NULL;
450void dump_send_ipi(void (*dump_ipi_callback)(void *))
451{
452	int i;
453	int cpu = smp_processor_id();
454
455	dump_ipi_function_ptr = dump_ipi_callback;
456	smp_mb();
457	for_each_online_cpu(i)
458		if (i != cpu)
459			mp_ops->send_ipi_single(i, SMP_DUMP);
460
461}
462EXPORT_SYMBOL(dump_send_ipi);
463#endif
464
465#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
466
467static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
468static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
469
470void tick_broadcast(const struct cpumask *mask)
471{
472	atomic_t *count;
473	struct call_single_data *csd;
474	int cpu;
475
476	for_each_cpu(cpu, mask) {
477		count = &per_cpu(tick_broadcast_count, cpu);
478		csd = &per_cpu(tick_broadcast_csd, cpu);
479
480		if (atomic_inc_return(count) == 1)
481			smp_call_function_single_async(cpu, csd);
482	}
483}
484
485static void tick_broadcast_callee(void *info)
486{
487	int cpu = smp_processor_id();
488	tick_receive_broadcast();
489	atomic_set(&per_cpu(tick_broadcast_count, cpu), 0);
490}
491
492static int __init tick_broadcast_init(void)
493{
494	struct call_single_data *csd;
495	int cpu;
496
497	for (cpu = 0; cpu < NR_CPUS; cpu++) {
498		csd = &per_cpu(tick_broadcast_csd, cpu);
499		csd->func = tick_broadcast_callee;
500	}
501
502	return 0;
503}
504early_initcall(tick_broadcast_init);
505
506#endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */
507