1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
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
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/profile.h>
17 #include <linux/seq_file.h>
18 #include <linux/smp.h>
19 #include <linux/threads.h>
20 #include <linux/export.h>
21 #include <linux/syscore_ops.h>
22 #include <linux/time.h>
23 #include <linux/tracepoint.h>
24 #include <linux/sched/hotplug.h>
25 #include <linux/sched/task_stack.h>
29 #include <asm/loongson.h>
30 #include <asm/mmu_context.h>
32 #include <asm/processor.h>
33 #include <asm/setup.h>
36 int __cpu_number_map[NR_CPUS]; /* Map physical to logical */
37 EXPORT_SYMBOL(__cpu_number_map);
39 int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */
40 EXPORT_SYMBOL(__cpu_logical_map);
42 /* Representing the threads (siblings) of each logical CPU */
43 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
44 EXPORT_SYMBOL(cpu_sibling_map);
46 /* Representing the core map of multi-core chips of each logical CPU */
47 cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
48 EXPORT_SYMBOL(cpu_core_map);
50 static DECLARE_COMPLETION(cpu_starting);
51 static DECLARE_COMPLETION(cpu_running);
54 * A logcal cpu mask containing only one VPE per core to
55 * reduce the number of IPIs on large MT systems.
57 cpumask_t cpu_foreign_map[NR_CPUS] __read_mostly;
58 EXPORT_SYMBOL(cpu_foreign_map);
60 /* representing cpus for which sibling maps can be computed */
61 static cpumask_t cpu_sibling_setup_map;
63 /* representing cpus for which core maps can be computed */
64 static cpumask_t cpu_core_setup_map;
66 struct secondary_data cpuboot_data;
67 static DEFINE_PER_CPU(int, cpu_state);
74 static const char *ipi_types[NR_IPI] __tracepoint_string = {
75 [IPI_RESCHEDULE] = "Rescheduling interrupts",
76 [IPI_CALL_FUNCTION] = "Function call interrupts",
79 void show_ipi_list(struct seq_file *p, int prec)
83 for (i = 0; i < NR_IPI; i++) {
84 seq_printf(p, "%*s%u:%s", prec - 1, "IPI", i, prec >= 4 ? " " : "");
85 for_each_online_cpu(cpu)
86 seq_printf(p, "%10u ", per_cpu(irq_stat, cpu).ipi_irqs[i]);
87 seq_printf(p, " LoongArch %d %s\n", i + 1, ipi_types[i]);
91 /* Send mailbox buffer via Mail_Send */
92 static void csr_mail_send(uint64_t data, int cpu, int mailbox)
96 /* Send high 32 bits */
97 val = IOCSR_MBUF_SEND_BLOCKING;
98 val |= (IOCSR_MBUF_SEND_BOX_HI(mailbox) << IOCSR_MBUF_SEND_BOX_SHIFT);
99 val |= (cpu << IOCSR_MBUF_SEND_CPU_SHIFT);
100 val |= (data & IOCSR_MBUF_SEND_H32_MASK);
101 iocsr_write64(val, LOONGARCH_IOCSR_MBUF_SEND);
103 /* Send low 32 bits */
104 val = IOCSR_MBUF_SEND_BLOCKING;
105 val |= (IOCSR_MBUF_SEND_BOX_LO(mailbox) << IOCSR_MBUF_SEND_BOX_SHIFT);
106 val |= (cpu << IOCSR_MBUF_SEND_CPU_SHIFT);
107 val |= (data << IOCSR_MBUF_SEND_BUF_SHIFT);
108 iocsr_write64(val, LOONGARCH_IOCSR_MBUF_SEND);
111 static u32 ipi_read_clear(int cpu)
115 /* Load the ipi register to figure out what we're supposed to do */
116 action = iocsr_read32(LOONGARCH_IOCSR_IPI_STATUS);
117 /* Clear the ipi register to clear the interrupt */
118 iocsr_write32(action, LOONGARCH_IOCSR_IPI_CLEAR);
124 static void ipi_write_action(int cpu, u32 action)
126 unsigned int irq = 0;
128 while ((irq = ffs(action))) {
129 uint32_t val = IOCSR_IPI_SEND_BLOCKING;
132 val |= (cpu << IOCSR_IPI_SEND_CPU_SHIFT);
133 iocsr_write32(val, LOONGARCH_IOCSR_IPI_SEND);
134 action &= ~BIT(irq - 1);
138 void loongson_send_ipi_single(int cpu, unsigned int action)
140 ipi_write_action(cpu_logical_map(cpu), (u32)action);
143 void loongson_send_ipi_mask(const struct cpumask *mask, unsigned int action)
147 for_each_cpu(i, mask)
148 ipi_write_action(cpu_logical_map(i), (u32)action);
152 * This function sends a 'reschedule' IPI to another CPU.
153 * it goes straight through and wastes no time serializing
154 * anything. Worst case is that we lose a reschedule ...
156 void arch_smp_send_reschedule(int cpu)
158 loongson_send_ipi_single(cpu, SMP_RESCHEDULE);
160 EXPORT_SYMBOL_GPL(arch_smp_send_reschedule);
162 irqreturn_t loongson_ipi_interrupt(int irq, void *dev)
165 unsigned int cpu = smp_processor_id();
167 action = ipi_read_clear(cpu_logical_map(cpu));
169 if (action & SMP_RESCHEDULE) {
171 per_cpu(irq_stat, cpu).ipi_irqs[IPI_RESCHEDULE]++;
174 if (action & SMP_CALL_FUNCTION) {
175 generic_smp_call_function_interrupt();
176 per_cpu(irq_stat, cpu).ipi_irqs[IPI_CALL_FUNCTION]++;
182 static void __init fdt_smp_setup(void)
185 unsigned int cpu, cpuid;
186 struct device_node *node = NULL;
188 for_each_of_cpu_node(node) {
189 if (!of_device_is_available(node))
192 cpuid = of_get_cpu_hwid(node, 0);
193 if (cpuid >= nr_cpu_ids)
196 if (cpuid == loongson_sysconf.boot_cpu_id) {
200 cpu = cpumask_next_zero(-1, cpu_present_mask);
204 set_cpu_possible(cpu, true);
205 set_cpu_present(cpu, true);
206 __cpu_number_map[cpuid] = cpu;
207 __cpu_logical_map[cpu] = cpuid;
210 loongson_sysconf.nr_cpus = num_processors;
211 set_bit(0, &(loongson_sysconf.cores_io_master));
215 void __init loongson_smp_setup(void)
219 cpu_data[0].core = cpu_logical_map(0) % loongson_sysconf.cores_per_package;
220 cpu_data[0].package = cpu_logical_map(0) / loongson_sysconf.cores_per_package;
222 iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
223 pr_info("Detected %i available CPU(s)\n", loongson_sysconf.nr_cpus);
226 void __init loongson_prepare_cpus(unsigned int max_cpus)
230 parse_acpi_topology();
232 for (i = 0; i < loongson_sysconf.nr_cpus; i++) {
233 set_cpu_present(i, true);
234 csr_mail_send(0, __cpu_logical_map[i], 0);
235 cpu_data[i].global_id = __cpu_logical_map[i];
238 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
242 * Setup the PC, SP, and TP of a secondary processor and start it running!
244 void loongson_boot_secondary(int cpu, struct task_struct *idle)
248 pr_info("Booting CPU#%d...\n", cpu);
250 entry = __pa_symbol((unsigned long)&smpboot_entry);
251 cpuboot_data.stack = (unsigned long)__KSTK_TOS(idle);
252 cpuboot_data.thread_info = (unsigned long)task_thread_info(idle);
254 csr_mail_send(entry, cpu_logical_map(cpu), 0);
256 loongson_send_ipi_single(cpu, SMP_BOOT_CPU);
260 * SMP init and finish on secondary CPUs
262 void loongson_init_secondary(void)
264 unsigned int cpu = smp_processor_id();
265 unsigned int imask = ECFGF_IP0 | ECFGF_IP1 | ECFGF_IP2 |
266 ECFGF_IPI | ECFGF_PMC | ECFGF_TIMER;
268 change_csr_ecfg(ECFG0_IM, imask);
270 iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
275 per_cpu(cpu_state, cpu) = CPU_ONLINE;
276 cpu_data[cpu].package =
277 cpu_logical_map(cpu) / loongson_sysconf.cores_per_package;
278 cpu_data[cpu].core = pptt_enabled ? cpu_data[cpu].core :
279 cpu_logical_map(cpu) % loongson_sysconf.cores_per_package;
282 void loongson_smp_finish(void)
285 iocsr_write64(0, LOONGARCH_IOCSR_MBUF0);
286 pr_info("CPU#%d finished\n", smp_processor_id());
289 #ifdef CONFIG_HOTPLUG_CPU
291 int loongson_cpu_disable(void)
294 unsigned int cpu = smp_processor_id();
300 numa_remove_cpu(cpu);
302 set_cpu_online(cpu, false);
303 calculate_cpu_foreign_map();
304 local_irq_save(flags);
305 irq_migrate_all_off_this_cpu();
306 clear_csr_ecfg(ECFG0_IM);
307 local_irq_restore(flags);
308 local_flush_tlb_all();
313 void loongson_cpu_die(unsigned int cpu)
315 while (per_cpu(cpu_state, cpu) != CPU_DEAD)
321 void __noreturn arch_cpu_idle_dead(void)
323 register uint64_t addr;
324 register void (*init_fn)(void);
328 set_csr_ecfg(ECFGF_IPI);
329 __this_cpu_write(cpu_state, CPU_DEAD);
333 __asm__ __volatile__("idle 0\n\t");
334 addr = iocsr_read64(LOONGARCH_IOCSR_MBUF0);
337 init_fn = (void *)TO_CACHE(addr);
338 iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_CLEAR);
351 static int loongson_ipi_suspend(void)
356 static void loongson_ipi_resume(void)
358 iocsr_write32(0xffffffff, LOONGARCH_IOCSR_IPI_EN);
361 static struct syscore_ops loongson_ipi_syscore_ops = {
362 .resume = loongson_ipi_resume,
363 .suspend = loongson_ipi_suspend,
367 * Enable boot cpu ipi before enabling nonboot cpus
368 * during syscore_resume.
370 static int __init ipi_pm_init(void)
372 register_syscore_ops(&loongson_ipi_syscore_ops);
376 core_initcall(ipi_pm_init);
379 static inline void set_cpu_sibling_map(int cpu)
383 cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
385 for_each_cpu(i, &cpu_sibling_setup_map) {
386 if (cpus_are_siblings(cpu, i)) {
387 cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
388 cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
393 static inline void set_cpu_core_map(int cpu)
397 cpumask_set_cpu(cpu, &cpu_core_setup_map);
399 for_each_cpu(i, &cpu_core_setup_map) {
400 if (cpu_data[cpu].package == cpu_data[i].package) {
401 cpumask_set_cpu(i, &cpu_core_map[cpu]);
402 cpumask_set_cpu(cpu, &cpu_core_map[i]);
408 * Calculate a new cpu_foreign_map mask whenever a
409 * new cpu appears or disappears.
411 void calculate_cpu_foreign_map(void)
413 int i, k, core_present;
414 cpumask_t temp_foreign_map;
416 /* Re-calculate the mask */
417 cpumask_clear(&temp_foreign_map);
418 for_each_online_cpu(i) {
420 for_each_cpu(k, &temp_foreign_map)
421 if (cpus_are_siblings(i, k))
424 cpumask_set_cpu(i, &temp_foreign_map);
427 for_each_online_cpu(i)
428 cpumask_andnot(&cpu_foreign_map[i],
429 &temp_foreign_map, &cpu_sibling_map[i]);
432 /* Preload SMP state for boot cpu */
433 void smp_prepare_boot_cpu(void)
435 unsigned int cpu, node, rr_node;
437 set_cpu_possible(0, true);
438 set_cpu_online(0, true);
439 set_my_cpu_offset(per_cpu_offset(0));
441 rr_node = first_node(node_online_map);
442 for_each_possible_cpu(cpu) {
443 node = early_cpu_to_node(cpu);
446 * The mapping between present cpus and nodes has been
447 * built during MADT and SRAT parsing.
449 * If possible cpus = present cpus here, early_cpu_to_node
450 * will return valid node.
452 * If possible cpus > present cpus here (e.g. some possible
453 * cpus will be added by cpu-hotplug later), for possible but
454 * not present cpus, early_cpu_to_node will return NUMA_NO_NODE,
455 * and we just map them to online nodes in round-robin way.
456 * Once hotplugged, new correct mapping will be built for them.
458 if (node != NUMA_NO_NODE)
459 set_cpu_numa_node(cpu, node);
461 set_cpu_numa_node(cpu, rr_node);
462 rr_node = next_node_in(rr_node, node_online_map);
467 /* called from main before smp_init() */
468 void __init smp_prepare_cpus(unsigned int max_cpus)
470 init_new_context(current, &init_mm);
471 current_thread_info()->cpu = 0;
472 loongson_prepare_cpus(max_cpus);
473 set_cpu_sibling_map(0);
475 calculate_cpu_foreign_map();
476 #ifndef CONFIG_HOTPLUG_CPU
477 init_cpu_present(cpu_possible_mask);
481 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
483 loongson_boot_secondary(cpu, tidle);
485 /* Wait for CPU to start and be ready to sync counters */
486 if (!wait_for_completion_timeout(&cpu_starting,
487 msecs_to_jiffies(5000))) {
488 pr_crit("CPU%u: failed to start\n", cpu);
492 /* Wait for CPU to finish startup & mark itself online before return */
493 wait_for_completion(&cpu_running);
499 * First C code run on the secondary CPUs after being started up by
502 asmlinkage void start_secondary(void)
507 cpu = smp_processor_id();
508 set_my_cpu_offset(per_cpu_offset(cpu));
511 constant_clockevent_init();
512 loongson_init_secondary();
514 set_cpu_sibling_map(cpu);
515 set_cpu_core_map(cpu);
517 notify_cpu_starting(cpu);
519 /* Notify boot CPU that we're starting */
520 complete(&cpu_starting);
522 /* The CPU is running, now mark it online */
523 set_cpu_online(cpu, true);
525 calculate_cpu_foreign_map();
528 * Notify boot CPU that we're up & online and it can safely return
531 complete(&cpu_running);
534 * irq will be enabled in loongson_smp_finish(), enabling it too
535 * early is dangerous.
537 WARN_ON_ONCE(!irqs_disabled());
538 loongson_smp_finish();
540 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
543 void __init smp_cpus_done(unsigned int max_cpus)
547 static void stop_this_cpu(void *dummy)
549 set_cpu_online(smp_processor_id(), false);
550 calculate_cpu_foreign_map();
555 void smp_send_stop(void)
557 smp_call_function(stop_this_cpu, NULL, 0);
560 #ifdef CONFIG_PROFILING
561 int setup_profiling_timer(unsigned int multiplier)
567 static void flush_tlb_all_ipi(void *info)
569 local_flush_tlb_all();
572 void flush_tlb_all(void)
574 on_each_cpu(flush_tlb_all_ipi, NULL, 1);
577 static void flush_tlb_mm_ipi(void *mm)
579 local_flush_tlb_mm((struct mm_struct *)mm);
582 void flush_tlb_mm(struct mm_struct *mm)
584 if (atomic_read(&mm->mm_users) == 0)
585 return; /* happens as a result of exit_mmap() */
589 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
590 on_each_cpu_mask(mm_cpumask(mm), flush_tlb_mm_ipi, mm, 1);
594 for_each_online_cpu(cpu) {
595 if (cpu != smp_processor_id() && cpu_context(cpu, mm))
596 cpu_context(cpu, mm) = 0;
598 local_flush_tlb_mm(mm);
604 struct flush_tlb_data {
605 struct vm_area_struct *vma;
610 static void flush_tlb_range_ipi(void *info)
612 struct flush_tlb_data *fd = info;
614 local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
617 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
619 struct mm_struct *mm = vma->vm_mm;
622 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
623 struct flush_tlb_data fd = {
629 on_each_cpu_mask(mm_cpumask(mm), flush_tlb_range_ipi, &fd, 1);
633 for_each_online_cpu(cpu) {
634 if (cpu != smp_processor_id() && cpu_context(cpu, mm))
635 cpu_context(cpu, mm) = 0;
637 local_flush_tlb_range(vma, start, end);
642 static void flush_tlb_kernel_range_ipi(void *info)
644 struct flush_tlb_data *fd = info;
646 local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
649 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
651 struct flush_tlb_data fd = {
656 on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
659 static void flush_tlb_page_ipi(void *info)
661 struct flush_tlb_data *fd = info;
663 local_flush_tlb_page(fd->vma, fd->addr1);
666 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
669 if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
670 struct flush_tlb_data fd = {
675 on_each_cpu_mask(mm_cpumask(vma->vm_mm), flush_tlb_page_ipi, &fd, 1);
679 for_each_online_cpu(cpu) {
680 if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
681 cpu_context(cpu, vma->vm_mm) = 0;
683 local_flush_tlb_page(vma, page);
687 EXPORT_SYMBOL(flush_tlb_page);
689 static void flush_tlb_one_ipi(void *info)
691 unsigned long vaddr = (unsigned long) info;
693 local_flush_tlb_one(vaddr);
696 void flush_tlb_one(unsigned long vaddr)
698 on_each_cpu(flush_tlb_one_ipi, (void *)vaddr, 1);
700 EXPORT_SYMBOL(flush_tlb_one);