xref: /linux/arch/loongarch/kernel/smp.c (revision 4f77c3462308c62ffe7129cc18b9ac937f44b5a5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
4  *
5  * Derived from MIPS:
6  * Copyright (C) 2000, 2001 Kanoj Sarcar
7  * Copyright (C) 2000, 2001 Ralf Baechle
8  * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
9  * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
10  */
11 #include <linux/acpi.h>
12 #include <linux/cpu.h>
13 #include <linux/cpumask.h>
14 #include <linux/init.h>
15 #include <linux/interrupt.h>
16 #include <linux/irq_work.h>
17 #include <linux/profile.h>
18 #include <linux/seq_file.h>
19 #include <linux/smp.h>
20 #include <linux/threads.h>
21 #include <linux/export.h>
22 #include <linux/syscore_ops.h>
23 #include <linux/time.h>
24 #include <linux/tracepoint.h>
25 #include <linux/sched/hotplug.h>
26 #include <linux/sched/task_stack.h>
27 
28 #include <asm/cpu.h>
29 #include <asm/idle.h>
30 #include <asm/loongson.h>
31 #include <asm/mmu_context.h>
32 #include <asm/numa.h>
33 #include <asm/paravirt.h>
34 #include <asm/processor.h>
35 #include <asm/setup.h>
36 #include <asm/time.h>
37 
38 int __cpu_number_map[NR_CPUS];   /* Map physical to logical */
39 EXPORT_SYMBOL(__cpu_number_map);
40 
41 int __cpu_logical_map[NR_CPUS];		/* Map logical to physical */
42 EXPORT_SYMBOL(__cpu_logical_map);
43 
44 /* Representing the threads (siblings) of each logical CPU */
45 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
46 EXPORT_SYMBOL(cpu_sibling_map);
47 
48 /* Representing the core map of multi-core chips of each logical CPU */
49 cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
50 EXPORT_SYMBOL(cpu_core_map);
51 
52 static DECLARE_COMPLETION(cpu_starting);
53 static DECLARE_COMPLETION(cpu_running);
54 
55 /*
56  * A logcal cpu mask containing only one VPE per core to
57  * reduce the number of IPIs on large MT systems.
58  */
59 cpumask_t cpu_foreign_map[NR_CPUS] __read_mostly;
60 EXPORT_SYMBOL(cpu_foreign_map);
61 
62 /* representing cpus for which sibling maps can be computed */
63 static cpumask_t cpu_sibling_setup_map;
64 
65 /* representing cpus for which core maps can be computed */
66 static cpumask_t cpu_core_setup_map;
67 
68 struct secondary_data cpuboot_data;
69 static DEFINE_PER_CPU(int, cpu_state);
70 
71 static const char *ipi_types[NR_IPI] __tracepoint_string = {
72 	[IPI_RESCHEDULE] = "Rescheduling interrupts",
73 	[IPI_CALL_FUNCTION] = "Function call interrupts",
74 	[IPI_IRQ_WORK] = "IRQ work interrupts",
75 };
76 
77 void show_ipi_list(struct seq_file *p, int prec)
78 {
79 	unsigned int cpu, i;
80 
81 	for (i = 0; i < NR_IPI; i++) {
82 		seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i, prec >= 4 ? " " : "");
83 		for_each_online_cpu(cpu)
84 			seq_printf(p, "%10u ", per_cpu(irq_stat, cpu).ipi_irqs[i]);
85 		seq_printf(p, " LoongArch  %d  %s\n", i + 1, ipi_types[i]);
86 	}
87 }
88 
89 static inline void set_cpu_core_map(int cpu)
90 {
91 	int i;
92 
93 	cpumask_set_cpu(cpu, &cpu_core_setup_map);
94 
95 	for_each_cpu(i, &cpu_core_setup_map) {
96 		if (cpu_data[cpu].package == cpu_data[i].package) {
97 			cpumask_set_cpu(i, &cpu_core_map[cpu]);
98 			cpumask_set_cpu(cpu, &cpu_core_map[i]);
99 		}
100 	}
101 }
102 
103 static inline void set_cpu_sibling_map(int cpu)
104 {
105 	int i;
106 
107 	cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
108 
109 	for_each_cpu(i, &cpu_sibling_setup_map) {
110 		if (cpus_are_siblings(cpu, i)) {
111 			cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
112 			cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
113 		}
114 	}
115 }
116 
117 static inline void clear_cpu_sibling_map(int cpu)
118 {
119 	int i;
120 
121 	for_each_cpu(i, &cpu_sibling_setup_map) {
122 		if (cpus_are_siblings(cpu, i)) {
123 			cpumask_clear_cpu(i, &cpu_sibling_map[cpu]);
124 			cpumask_clear_cpu(cpu, &cpu_sibling_map[i]);
125 		}
126 	}
127 
128 	cpumask_clear_cpu(cpu, &cpu_sibling_setup_map);
129 }
130 
131 /*
132  * Calculate a new cpu_foreign_map mask whenever a
133  * new cpu appears or disappears.
134  */
135 void calculate_cpu_foreign_map(void)
136 {
137 	int i, k, core_present;
138 	cpumask_t temp_foreign_map;
139 
140 	/* Re-calculate the mask */
141 	cpumask_clear(&temp_foreign_map);
142 	for_each_online_cpu(i) {
143 		core_present = 0;
144 		for_each_cpu(k, &temp_foreign_map)
145 			if (cpus_are_siblings(i, k))
146 				core_present = 1;
147 		if (!core_present)
148 			cpumask_set_cpu(i, &temp_foreign_map);
149 	}
150 
151 	for_each_online_cpu(i)
152 		cpumask_andnot(&cpu_foreign_map[i],
153 			       &temp_foreign_map, &cpu_sibling_map[i]);
154 }
155 
156 /* Send mailbox buffer via Mail_Send */
157 static void csr_mail_send(uint64_t data, int cpu, int mailbox)
158 {
159 	uint64_t val;
160 
161 	/* Send high 32 bits */
162 	val = IOCSR_MBUF_SEND_BLOCKING;
163 	val |= (IOCSR_MBUF_SEND_BOX_HI(mailbox) << IOCSR_MBUF_SEND_BOX_SHIFT);
164 	val |= (cpu << IOCSR_MBUF_SEND_CPU_SHIFT);
165 	val |= (data & IOCSR_MBUF_SEND_H32_MASK);
166 	iocsr_write64(val, LOONGARCH_IOCSR_MBUF_SEND);
167 
168 	/* Send low 32 bits */
169 	val = IOCSR_MBUF_SEND_BLOCKING;
170 	val |= (IOCSR_MBUF_SEND_BOX_LO(mailbox) << IOCSR_MBUF_SEND_BOX_SHIFT);
171 	val |= (cpu << IOCSR_MBUF_SEND_CPU_SHIFT);
172 	val |= (data << IOCSR_MBUF_SEND_BUF_SHIFT);
173 	iocsr_write64(val, LOONGARCH_IOCSR_MBUF_SEND);
174 };
175 
176 static u32 ipi_read_clear(int cpu)
177 {
178 	u32 action;
179 
180 	/* Load the ipi register to figure out what we're supposed to do */
181 	action = iocsr_read32(LOONGARCH_IOCSR_IPI_STATUS);
182 	/* Clear the ipi register to clear the interrupt */
183 	iocsr_write32(action, LOONGARCH_IOCSR_IPI_CLEAR);
184 	wbflush();
185 
186 	return action;
187 }
188 
189 static void ipi_write_action(int cpu, u32 action)
190 {
191 	uint32_t val;
192 
193 	val = IOCSR_IPI_SEND_BLOCKING | action;
194 	val |= (cpu << IOCSR_IPI_SEND_CPU_SHIFT);
195 	iocsr_write32(val, LOONGARCH_IOCSR_IPI_SEND);
196 }
197 
198 static void loongson_send_ipi_single(int cpu, unsigned int action)
199 {
200 	ipi_write_action(cpu_logical_map(cpu), (u32)action);
201 }
202 
203 static void loongson_send_ipi_mask(const struct cpumask *mask, unsigned int action)
204 {
205 	unsigned int i;
206 
207 	for_each_cpu(i, mask)
208 		ipi_write_action(cpu_logical_map(i), (u32)action);
209 }
210 
211 /*
212  * This function sends a 'reschedule' IPI to another CPU.
213  * it goes straight through and wastes no time serializing
214  * anything. Worst case is that we lose a reschedule ...
215  */
216 void arch_smp_send_reschedule(int cpu)
217 {
218 	mp_ops.send_ipi_single(cpu, ACTION_RESCHEDULE);
219 }
220 EXPORT_SYMBOL_GPL(arch_smp_send_reschedule);
221 
222 #ifdef CONFIG_IRQ_WORK
223 void arch_irq_work_raise(void)
224 {
225 	mp_ops.send_ipi_single(smp_processor_id(), ACTION_IRQ_WORK);
226 }
227 #endif
228 
229 static irqreturn_t loongson_ipi_interrupt(int irq, void *dev)
230 {
231 	unsigned int action;
232 	unsigned int cpu = smp_processor_id();
233 
234 	action = ipi_read_clear(cpu_logical_map(cpu));
235 
236 	if (action & SMP_RESCHEDULE) {
237 		scheduler_ipi();
238 		per_cpu(irq_stat, cpu).ipi_irqs[IPI_RESCHEDULE]++;
239 	}
240 
241 	if (action & SMP_CALL_FUNCTION) {
242 		generic_smp_call_function_interrupt();
243 		per_cpu(irq_stat, cpu).ipi_irqs[IPI_CALL_FUNCTION]++;
244 	}
245 
246 	if (action & SMP_IRQ_WORK) {
247 		irq_work_run();
248 		per_cpu(irq_stat, cpu).ipi_irqs[IPI_IRQ_WORK]++;
249 	}
250 
251 	return IRQ_HANDLED;
252 }
253 
254 static void loongson_init_ipi(void)
255 {
256 	int r, ipi_irq;
257 
258 	ipi_irq = get_percpu_irq(INT_IPI);
259 	if (ipi_irq < 0)
260 		panic("IPI IRQ mapping failed\n");
261 
262 	irq_set_percpu_devid(ipi_irq);
263 	r = request_percpu_irq(ipi_irq, loongson_ipi_interrupt, "IPI", &irq_stat);
264 	if (r < 0)
265 		panic("IPI IRQ request failed\n");
266 }
267 
268 struct smp_ops mp_ops = {
269 	.init_ipi		= loongson_init_ipi,
270 	.send_ipi_single	= loongson_send_ipi_single,
271 	.send_ipi_mask		= loongson_send_ipi_mask,
272 };
273 
274 static void __init fdt_smp_setup(void)
275 {
276 #ifdef CONFIG_OF
277 	unsigned int cpu, cpuid;
278 	struct device_node *node = NULL;
279 
280 	for_each_of_cpu_node(node) {
281 		if (!of_device_is_available(node))
282 			continue;
283 
284 		cpuid = of_get_cpu_hwid(node, 0);
285 		if (cpuid >= nr_cpu_ids)
286 			continue;
287 
288 		if (cpuid == loongson_sysconf.boot_cpu_id)
289 			cpu = 0;
290 		else
291 			cpu = find_first_zero_bit(cpumask_bits(cpu_present_mask), NR_CPUS);
292 
293 		num_processors++;
294 		set_cpu_possible(cpu, true);
295 		set_cpu_present(cpu, true);
296 		__cpu_number_map[cpuid] = cpu;
297 		__cpu_logical_map[cpu] = cpuid;
298 
299 		early_numa_add_cpu(cpu, 0);
300 		set_cpuid_to_node(cpuid, 0);
301 	}
302 
303 	loongson_sysconf.nr_cpus = num_processors;
304 	set_bit(0, loongson_sysconf.cores_io_master);
305 #endif
306 }
307 
308 void __init loongson_smp_setup(void)
309 {
310 	fdt_smp_setup();
311 
312 	if (loongson_sysconf.cores_per_package == 0)
313 		loongson_sysconf.cores_per_package = num_processors;
314 
315 	cpu_data[0].core = cpu_logical_map(0) % loongson_sysconf.cores_per_package;
316 	cpu_data[0].package = cpu_logical_map(0) / loongson_sysconf.cores_per_package;
317 
318 	pv_ipi_init();
319 	iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
320 	pr_info("Detected %i available CPU(s)\n", loongson_sysconf.nr_cpus);
321 }
322 
323 void __init loongson_prepare_cpus(unsigned int max_cpus)
324 {
325 	int i = 0;
326 
327 	parse_acpi_topology();
328 
329 	for (i = 0; i < loongson_sysconf.nr_cpus; i++) {
330 		set_cpu_present(i, true);
331 		csr_mail_send(0, __cpu_logical_map[i], 0);
332 		cpu_data[i].global_id = __cpu_logical_map[i];
333 	}
334 
335 	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
336 }
337 
338 /*
339  * Setup the PC, SP, and TP of a secondary processor and start it running!
340  */
341 void loongson_boot_secondary(int cpu, struct task_struct *idle)
342 {
343 	unsigned long entry;
344 
345 	pr_info("Booting CPU#%d...\n", cpu);
346 
347 	entry = __pa_symbol((unsigned long)&smpboot_entry);
348 	cpuboot_data.stack = (unsigned long)__KSTK_TOS(idle);
349 	cpuboot_data.thread_info = (unsigned long)task_thread_info(idle);
350 
351 	csr_mail_send(entry, cpu_logical_map(cpu), 0);
352 
353 	loongson_send_ipi_single(cpu, ACTION_BOOT_CPU);
354 }
355 
356 /*
357  * SMP init and finish on secondary CPUs
358  */
359 void loongson_init_secondary(void)
360 {
361 	unsigned int cpu = smp_processor_id();
362 	unsigned int imask = ECFGF_IP0 | ECFGF_IP1 | ECFGF_IP2 |
363 			     ECFGF_IPI | ECFGF_PMC | ECFGF_TIMER | ECFGF_SIP0;
364 
365 	change_csr_ecfg(ECFG0_IM, imask);
366 
367 	iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
368 
369 #ifdef CONFIG_NUMA
370 	numa_add_cpu(cpu);
371 #endif
372 	per_cpu(cpu_state, cpu) = CPU_ONLINE;
373 	cpu_data[cpu].package =
374 		     cpu_logical_map(cpu) / loongson_sysconf.cores_per_package;
375 	cpu_data[cpu].core = pptt_enabled ? cpu_data[cpu].core :
376 		     cpu_logical_map(cpu) % loongson_sysconf.cores_per_package;
377 }
378 
379 void loongson_smp_finish(void)
380 {
381 	local_irq_enable();
382 	iocsr_write64(0, LOONGARCH_IOCSR_MBUF0);
383 	pr_info("CPU#%d finished\n", smp_processor_id());
384 }
385 
386 #ifdef CONFIG_HOTPLUG_CPU
387 
388 int loongson_cpu_disable(void)
389 {
390 	unsigned long flags;
391 	unsigned int cpu = smp_processor_id();
392 
393 	if (io_master(cpu))
394 		return -EBUSY;
395 
396 #ifdef CONFIG_NUMA
397 	numa_remove_cpu(cpu);
398 #endif
399 	set_cpu_online(cpu, false);
400 	clear_cpu_sibling_map(cpu);
401 	calculate_cpu_foreign_map();
402 	local_irq_save(flags);
403 	irq_migrate_all_off_this_cpu();
404 	clear_csr_ecfg(ECFG0_IM);
405 	local_irq_restore(flags);
406 	local_flush_tlb_all();
407 
408 	return 0;
409 }
410 
411 void loongson_cpu_die(unsigned int cpu)
412 {
413 	while (per_cpu(cpu_state, cpu) != CPU_DEAD)
414 		cpu_relax();
415 
416 	mb();
417 }
418 
419 void __noreturn arch_cpu_idle_dead(void)
420 {
421 	register uint64_t addr;
422 	register void (*init_fn)(void);
423 
424 	idle_task_exit();
425 	local_irq_enable();
426 	set_csr_ecfg(ECFGF_IPI);
427 	__this_cpu_write(cpu_state, CPU_DEAD);
428 
429 	__smp_mb();
430 	do {
431 		__asm__ __volatile__("idle 0\n\t");
432 		addr = iocsr_read64(LOONGARCH_IOCSR_MBUF0);
433 	} while (addr == 0);
434 
435 	local_irq_disable();
436 	init_fn = (void *)TO_CACHE(addr);
437 	iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_CLEAR);
438 
439 	init_fn();
440 	BUG();
441 }
442 
443 #endif
444 
445 /*
446  * Power management
447  */
448 #ifdef CONFIG_PM
449 
450 static int loongson_ipi_suspend(void)
451 {
452 	return 0;
453 }
454 
455 static void loongson_ipi_resume(void)
456 {
457 	iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
458 }
459 
460 static struct syscore_ops loongson_ipi_syscore_ops = {
461 	.resume         = loongson_ipi_resume,
462 	.suspend        = loongson_ipi_suspend,
463 };
464 
465 /*
466  * Enable boot cpu ipi before enabling nonboot cpus
467  * during syscore_resume.
468  */
469 static int __init ipi_pm_init(void)
470 {
471 	register_syscore_ops(&loongson_ipi_syscore_ops);
472 	return 0;
473 }
474 
475 core_initcall(ipi_pm_init);
476 #endif
477 
478 /* Preload SMP state for boot cpu */
479 void smp_prepare_boot_cpu(void)
480 {
481 	unsigned int cpu, node, rr_node;
482 
483 	set_cpu_possible(0, true);
484 	set_cpu_online(0, true);
485 	set_my_cpu_offset(per_cpu_offset(0));
486 	numa_add_cpu(0);
487 
488 	rr_node = first_node(node_online_map);
489 	for_each_possible_cpu(cpu) {
490 		node = early_cpu_to_node(cpu);
491 
492 		/*
493 		 * The mapping between present cpus and nodes has been
494 		 * built during MADT and SRAT parsing.
495 		 *
496 		 * If possible cpus = present cpus here, early_cpu_to_node
497 		 * will return valid node.
498 		 *
499 		 * If possible cpus > present cpus here (e.g. some possible
500 		 * cpus will be added by cpu-hotplug later), for possible but
501 		 * not present cpus, early_cpu_to_node will return NUMA_NO_NODE,
502 		 * and we just map them to online nodes in round-robin way.
503 		 * Once hotplugged, new correct mapping will be built for them.
504 		 */
505 		if (node != NUMA_NO_NODE)
506 			set_cpu_numa_node(cpu, node);
507 		else {
508 			set_cpu_numa_node(cpu, rr_node);
509 			rr_node = next_node_in(rr_node, node_online_map);
510 		}
511 	}
512 }
513 
514 /* called from main before smp_init() */
515 void __init smp_prepare_cpus(unsigned int max_cpus)
516 {
517 	init_new_context(current, &init_mm);
518 	current_thread_info()->cpu = 0;
519 	loongson_prepare_cpus(max_cpus);
520 	set_cpu_sibling_map(0);
521 	set_cpu_core_map(0);
522 	calculate_cpu_foreign_map();
523 #ifndef CONFIG_HOTPLUG_CPU
524 	init_cpu_present(cpu_possible_mask);
525 #endif
526 }
527 
528 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
529 {
530 	loongson_boot_secondary(cpu, tidle);
531 
532 	/* Wait for CPU to start and be ready to sync counters */
533 	if (!wait_for_completion_timeout(&cpu_starting,
534 					 msecs_to_jiffies(5000))) {
535 		pr_crit("CPU%u: failed to start\n", cpu);
536 		return -EIO;
537 	}
538 
539 	/* Wait for CPU to finish startup & mark itself online before return */
540 	wait_for_completion(&cpu_running);
541 
542 	return 0;
543 }
544 
545 /*
546  * First C code run on the secondary CPUs after being started up by
547  * the master.
548  */
549 asmlinkage void start_secondary(void)
550 {
551 	unsigned int cpu;
552 
553 	sync_counter();
554 	cpu = raw_smp_processor_id();
555 	set_my_cpu_offset(per_cpu_offset(cpu));
556 
557 	cpu_probe();
558 	constant_clockevent_init();
559 	loongson_init_secondary();
560 
561 	set_cpu_sibling_map(cpu);
562 	set_cpu_core_map(cpu);
563 
564 	notify_cpu_starting(cpu);
565 
566 	/* Notify boot CPU that we're starting */
567 	complete(&cpu_starting);
568 
569 	/* The CPU is running, now mark it online */
570 	set_cpu_online(cpu, true);
571 
572 	calculate_cpu_foreign_map();
573 
574 	/*
575 	 * Notify boot CPU that we're up & online and it can safely return
576 	 * from __cpu_up()
577 	 */
578 	complete(&cpu_running);
579 
580 	/*
581 	 * irq will be enabled in loongson_smp_finish(), enabling it too
582 	 * early is dangerous.
583 	 */
584 	WARN_ON_ONCE(!irqs_disabled());
585 	loongson_smp_finish();
586 
587 	cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
588 }
589 
590 void __init smp_cpus_done(unsigned int max_cpus)
591 {
592 }
593 
594 static void stop_this_cpu(void *dummy)
595 {
596 	set_cpu_online(smp_processor_id(), false);
597 	calculate_cpu_foreign_map();
598 	local_irq_disable();
599 	while (true);
600 }
601 
602 void smp_send_stop(void)
603 {
604 	smp_call_function(stop_this_cpu, NULL, 0);
605 }
606 
607 #ifdef CONFIG_PROFILING
608 int setup_profiling_timer(unsigned int multiplier)
609 {
610 	return 0;
611 }
612 #endif
613 
614 static void flush_tlb_all_ipi(void *info)
615 {
616 	local_flush_tlb_all();
617 }
618 
619 void flush_tlb_all(void)
620 {
621 	on_each_cpu(flush_tlb_all_ipi, NULL, 1);
622 }
623 
624 static void flush_tlb_mm_ipi(void *mm)
625 {
626 	local_flush_tlb_mm((struct mm_struct *)mm);
627 }
628 
629 void flush_tlb_mm(struct mm_struct *mm)
630 {
631 	if (atomic_read(&mm->mm_users) == 0)
632 		return;		/* happens as a result of exit_mmap() */
633 
634 	preempt_disable();
635 
636 	if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
637 		on_each_cpu_mask(mm_cpumask(mm), flush_tlb_mm_ipi, mm, 1);
638 	} else {
639 		unsigned int cpu;
640 
641 		for_each_online_cpu(cpu) {
642 			if (cpu != smp_processor_id() && cpu_context(cpu, mm))
643 				cpu_context(cpu, mm) = 0;
644 		}
645 		local_flush_tlb_mm(mm);
646 	}
647 
648 	preempt_enable();
649 }
650 
651 struct flush_tlb_data {
652 	struct vm_area_struct *vma;
653 	unsigned long addr1;
654 	unsigned long addr2;
655 };
656 
657 static void flush_tlb_range_ipi(void *info)
658 {
659 	struct flush_tlb_data *fd = info;
660 
661 	local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
662 }
663 
664 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
665 {
666 	struct mm_struct *mm = vma->vm_mm;
667 
668 	preempt_disable();
669 	if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
670 		struct flush_tlb_data fd = {
671 			.vma = vma,
672 			.addr1 = start,
673 			.addr2 = end,
674 		};
675 
676 		on_each_cpu_mask(mm_cpumask(mm), flush_tlb_range_ipi, &fd, 1);
677 	} else {
678 		unsigned int cpu;
679 
680 		for_each_online_cpu(cpu) {
681 			if (cpu != smp_processor_id() && cpu_context(cpu, mm))
682 				cpu_context(cpu, mm) = 0;
683 		}
684 		local_flush_tlb_range(vma, start, end);
685 	}
686 	preempt_enable();
687 }
688 
689 static void flush_tlb_kernel_range_ipi(void *info)
690 {
691 	struct flush_tlb_data *fd = info;
692 
693 	local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
694 }
695 
696 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
697 {
698 	struct flush_tlb_data fd = {
699 		.addr1 = start,
700 		.addr2 = end,
701 	};
702 
703 	on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
704 }
705 
706 static void flush_tlb_page_ipi(void *info)
707 {
708 	struct flush_tlb_data *fd = info;
709 
710 	local_flush_tlb_page(fd->vma, fd->addr1);
711 }
712 
713 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
714 {
715 	preempt_disable();
716 	if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
717 		struct flush_tlb_data fd = {
718 			.vma = vma,
719 			.addr1 = page,
720 		};
721 
722 		on_each_cpu_mask(mm_cpumask(vma->vm_mm), flush_tlb_page_ipi, &fd, 1);
723 	} else {
724 		unsigned int cpu;
725 
726 		for_each_online_cpu(cpu) {
727 			if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
728 				cpu_context(cpu, vma->vm_mm) = 0;
729 		}
730 		local_flush_tlb_page(vma, page);
731 	}
732 	preempt_enable();
733 }
734 EXPORT_SYMBOL(flush_tlb_page);
735 
736 static void flush_tlb_one_ipi(void *info)
737 {
738 	unsigned long vaddr = (unsigned long) info;
739 
740 	local_flush_tlb_one(vaddr);
741 }
742 
743 void flush_tlb_one(unsigned long vaddr)
744 {
745 	on_each_cpu(flush_tlb_one_ipi, (void *)vaddr, 1);
746 }
747 EXPORT_SYMBOL(flush_tlb_one);
748