Linux irq_set_affinity中断亲和力设置与chip_irq_set

Linux irq_set_affinity中断亲和力设置与chip_irq_set中断亲和力IRQ Affinity是Linux内核将特定中断绑定到指定CPU核心执行的机制。通过设置irq_desc的affinity掩码可以控制中断路由到哪些CPU实现负载均衡和性能优化。核心接口是irq_set_affinity定义在kernel/irq/manage.c。irq_set_affinity函数的实现cint irq_set_affinity(unsigned int irq, const struct cpumask *cpumask){struct irq_desc *desc irq_to_desc(irq);unsigned long flags;int ret;if (!desc)return -EINVAL;raw_spin_lock_irqsave(desc-lock, flags);ret __irq_set_affinity_locked(irq_desc_get_irq_data(desc), cpumask, false);raw_spin_unlock_irqrestore(desc-lock, flags);return ret;}__irq_set_affinity_locked是实际执行设置的核心函数cint __irq_set_affinity_locked(struct irq_data *data, const struct cpumask *mask, bool force){struct irq_chip *chip irq_data_get_irq_chip(data);struct irq_desc *desc irq_data_to_desc(data);const struct cpumask *prog_mask;int ret -EINVAL;if (!chip || !chip-irq_set_affinity)goto out;if (!force !irq_can_set_affinity(data))goto out;prog_mask irq_data_get_affinity_mask(data);if (cpumask_equal(mask, prog_mask) cpumask_subset(prog_mask, cpu_online_mask))return 0;ret chip-irq_set_affinity(data, mask, force);if (!ret) {cpumask_copy(prog_mask, mask);irq_data_update_effective_affinity(data, mask);}out:return ret;}函数首先验证芯片支持irq_set_affinity回调且中断允许设置亲和力IRQ_NO_BALANCING标志未设置。如果新掩码与当前掩码相同且在线CPU子集未变化则直接返回0避免冗余的硬件操作。实际设置由chip-irq_set_affinity完成成功后将prog_mask复制为最新掩码。对于x86 IO-APICirq_set_affinity对应ioapic_set_affinitycstatic int ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force){struct irq_cfg *cfg data-chip_data;unsigned int dest;unsigned int vector;int cpu;cpu cpumask_first_and(mask, cpu_online_mask);if (cpu nr_cpu_ids)return -EINVAL;dest apic-calc_dest(cpu);vector cfg-vector;raw_spin_lock(ioapic_lock);__ioapic_write_entry(cfg-irq_2_pin, vector, dest, data-irq);raw_spin_unlock(ioapic_lock);cpumask_copy(irq_data_get_effective_affinity_mask(data), cpumask_of(cpu));return 0;}该实现首先从mask中选取第一个在线CPU通过calc_dest计算APIC destination ID将中断重定向到目标CPU的vector。IOAPIC的RTERedirection Table Entry中包含dest字段和vector字段硬件在中断产生时根据dest将中断发送到指定CPU的Local APIC。用户空间的亲和力设置通过/proc/irq/{irq}/smp_affinity文件暴露cstatic ssize_t smp_affinity_write(struct file *file, const char __user *buf,size_t count, loff_t *pos){struct irq_desc *desc file_inode(file)-i_private;cpumask_var_t new_value;int err;if (!zalloc_cpumask_var(new_value, GFP_KERNEL))return -ENOMEM;err cpumask_parse_user(buf, count, new_value);if (err)goto out;err irq_set_affinity(desc-irq_data.irq, new_value);if (!err)err count;out:free_cpumask_var(new_value);return err;}irqbalance守护进程周期性扫描/proc/interrupts统计各CPU中断计数动态调整smp_affinity。它计算每个中断在各CPU上的分布偏差若偏差超过阈值则调用irq_set_affinity将中断迁移到负载更低的CPU。默认策略是查找具有最低中断数的CPU核心。effective_affinity与requested_affinity的区别值得注意。requested_affinity是用户或内核请求的CPU掩码而effective_affinity是硬件实际路由的目标CPU。对于X86 MSI和MSI-X中断一个中断请求只能路由到一个CPU但用户请求可能包含多个CPU位此时内核选取第一个在线CPU写入硬件effective_affinity记录该单一CPUrequested_affinity保留原始多CPU掩码。中断迁移irq_migrate_all_off_this_cpu在CPU热拔插时被调用将即将离线的CPU上的所有中断迁移到其他在线CPUcvoid irq_migrate_all_off_this_cpu(void){unsigned int irq;struct irq_desc *desc;for_each_irq_desc(irq, desc) {bool affinity_broken false;const struct cpumask *affinity;affinity irq_data_get_affinity_mask(desc-irq_data);if (!cpumask_test_cpu(smp_processor_id(), affinity))continue;cpumask_clear_cpu(smp_processor_id(), affinity);if (cpumask_empty(affinity))cpumask_setall(affinity);__irq_set_affinity_locked(desc-irq_data, affinity, true);}}该函数在CPU_DOWN_PREPARE阶段执行确保所有中断被重新分配避免离线CPU无法处理中断导致中断丢失。调度域层面的中断负载均衡由内核线程irqbalance或内核参数irqaffinity控制。内核启动参数irqaffinity0-3允许在启动阶段指定默认亲和掩码通过init_irq_default_affinity设置到desc-irq_data.affinity。numa_node字段辅助将中断分配给与设备所在NUMA节点相同的CPU减少跨节点内存访问延迟。