Linux 在每个 CPU 上会创建一个 ksoftirqd 内核线程。
softirqs 是在 Linux 内核编译时就确定好的，例如网络收包对应的 NET_RX_SOFTIRQ 软中断。因此是一种静态机制。如果想加一种新 softirq 类型，就需要修改并重新编译内核。
代码语言：javascript复制heidsoft@heidsoft-dev:~$ systemd-cgls -k | grep kworker
├─ 8 [kworker/0:0H-events_highpri]
├─ 24 [kworker/1:0H-events_highpri]
├─ 91 [kworker/0:1H-kblockd]
├─ 126 [kworker/u65:0]
├─ 155 [kworker/1:1H-kblockd]
├─164504 [kworker/1:3-events]
├─166487 [kworker/u64:2-events_unbound]
├─166610 [kworker/0:1-events]
├─167421 [kworker/u64:0-events_unbound]
├─167437 [kworker/0:0-events]
├─167685 [kworker/1:1-events]
├─168449 [kworker/u64:1]
├─168597 [kworker/0:2-kec_query]heidsoft@heidsoft-dev:~$ cat /proc/interrupts
CPU0 CPU1
0: 7 0 IO-APIC 2-edge timer
1: 37 0 IO-APIC 1-edge i8042
7: 0 0 IO-APIC 7-edge parport0
8: 0 1 IO-APIC 8-edge rtc0
9: 0 176583 IO-APIC 9-fasteoi acpi
12: 0 144 IO-APIC 12-edge i8042
14: 0 0 IO-APIC 14-edge ata_piix
15: 0 0 IO-APIC 15-edge ata_piix
18: 38262 10860 IO-APIC 18-fasteoi uhci_hcd:usb2, ioc0
19: 136 61 IO-APIC 19-fasteoi ehci_hcd:usb1
22: 0 0 IO-APIC 22-fasteoi virtio1
24: 4520 0 PCI-MSI 49152-edge prl_tg
25: 201 109 PCI-MSI 81920-edge virtio0-config
26: 20479 411330 PCI-MSI 81921-edge virtio0-input.0
27: 471647 51205 PCI-MSI 81922-edge virtio0-output.0
28: 32 0 PCI-MSI 487424-edge xhci_hcd
29: 17040 15240 PCI-MSI 512000-edge ahci[0000:00:1f.2]
30: 0 1 PCI-MSI 524288-edge virtio2-config
31: 1131 1433 PCI-MSI 524289-edge virtio2-virtqueues
32: 0 4734 PCI-MSI 516096-edge snd_hda_intel:card0
NMI: 0 0 Non-maskable interrupts
LOC: 2110890 1998553 Local timer interrupts
SPU: 0 0 Spurious interrupts
PMI: 0 0 Performance monitoring interrupts
IWI: 0 0 IRQ work interrupts
RTR: 0 0 APIC ICR read retries
RES: 94259 90183 Rescheduling interrupts
CAL: 1563802 1451446 Function call interrupts
TLB: 1119 1363 TLB shootdowns
TRM: 0 0 Thermal event interrupts
THR: 0 0 Threshold APIC interrupts
DFR: 0 0 Deferred Error APIC interrupts
MCE: 0 0 Machine check exceptions
MCP: 210 210 Machine check polls
ERR: 0
MIS: 0
PIN: 0 0 Posted-interrupt notification event
NPI: 0 0 Nested posted-interrupt event
PIW: 0 0 Posted-interrupt wakeup event
heidsoft@heidsoft-dev:~$/research/linux-5.15.4/net/core/dev.c
注册网卡收发包（RX/TX）软中断处理函数
open_softirq(NET_TX_SOFTIRQ, net_tx_action);
open_softirq(NET_RX_SOFTIRQ, net_rx_action);
代码语言：javascript复制static int __init net_dev_init(void)
{
int i, rc = -ENOMEM;
BUG_ON(!dev_boot_phase);
if (dev_proc_init())
goto out;
if (netdev_kobject_init())
goto out;
INIT_LIST_HEAD(&ptype_all);
for (i = 0; i < PTYPE_HASH_SIZE; i )
INIT_LIST_HEAD(&ptype_base[i]);
INIT_LIST_HEAD(&offload_base);
if (register_pernet_subsys(&netdev_net_ops))
goto out;
/*
* Initialise the packet receive queues.
*/
for_each_possible_cpu(i) {
struct work_struct *flush = per_cpu_ptr(&flush_works, i);
struct softnet_data *sd = &per_cpu(softnet_data, i);
INIT_WORK(flush, flush_backlog);
skb_queue_head_init(&sd->input_pkt_queue);
skb_queue_head_init(&sd->process_queue);
#ifdef CONFIG_XFRM_OFFLOAD
skb_queue_head_init(&sd->xfrm_backlog);
#endif
INIT_LIST_HEAD(&sd->poll_list);
sd->output_queue_tailp = &sd->output_queue;
#ifdef CONFIG_RPS
INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
sd->cpu = i;
#endif
init_gro_hash(&sd->backlog);
sd->backlog.poll = process_backlog;
sd->backlog.weight = weight_p;
}
dev_boot_phase = 0;
/* The loopback device is special if any other network devices
* is present in a network namespace the loopback device must
* be present. Since we now dynamically allocate and free the
* loopback device ensure this invariant is maintained by
* keeping the loopback device as the first device on the
* list of network devices. Ensuring the loopback devices
* is the first device that appears and the last network device
* that disappears.
*/
if (register_pernet_device(&loopback_net_ops))
goto out;
if (register_pernet_device(&default_device_ops))
goto out;
open_softirq(NET_TX_SOFTIRQ, net_tx_action);
open_softirq(NET_RX_SOFTIRQ, net_rx_action);
rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
NULL, dev_cpu_dead);
WARN_ON(rc < 0);
rc = 0;
out:
return rc;
}heidsoft@heidsoft-dev:~$ sudo perf record -a
> -e irq:irq_handler_entry,irq:irq_handler_exit
> -e irq:softirq_entry --filter="vec == 3"
> -e irq:softirq_exit --filter="vec == 3"
> -e napi:napi_poll
> -- sleep 1
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.725 MB perf.data (20 samples) ]
heidsoft@heidsoft-dev:~$ sudo perf script
swapper 0 [000] 67144.378134: irq:irq_handler_entry: irq=18 name=uhci_hcd:usb2
swapper 0 [000] 67144.378145: irq:irq_handler_exit: irq=18 ret=unhandled
swapper 0 [000] 67144.378146: irq:irq_handler_entry: irq=18 name=ioc0
swapper 0 [000] 67144.378167: irq:irq_handler_exit: irq=18 ret=handled
swapper 0 [000] 67144.378685: irq:irq_handler_entry: irq=18 name=uhci_hcd:usb2
swapper 0 [000] 67144.378692: irq:irq_handler_exit: irq=18 ret=unhandled
swapper 0 [000] 67144.378692: irq:irq_handler_entry: irq=18 name=ioc0
swapper 0 [000] 67144.378710: irq:irq_handler_exit: irq=18 ret=handled
swapper 0 [001] 67144.512867: irq:irq_handler_entry: irq=26 name=virtio0-input.0
swapper 0 [001] 67144.512870: irq:irq_handler_exit: irq=26 ret=handled
swapper 0 [001] 67144.512873: irq:softirq_entry: vec=3 [action=NET_RX]
swapper 0 [001] 67144.514142: irq:irq_handler_entry: irq=27 name=virtio0-output.0
swapper 0 [001] 67144.514144: irq:irq_handler_exit: irq=27 ret=handled
swapper 0 [001] 67144.514146: napi:napi_poll: napi poll on napi struct 0xffff8dfa>
swapper 0 [001] 67144.514147: irq:softirq_exit: vec=3 [action=NET_RX]
swapper 0 [001] 67144.514148: irq:softirq_entry: vec=3 [action=NET_RX]
swapper 0 [001] 67144.514150: napi:napi_poll: napi poll on napi struct 0xffff8dfa>
swapper 0 [001] 67144.514151: irq:softirq_exit: vec=3 [action=NET_RX]
swapper 0 [000] 67144.519984: irq:irq_handler_entry: irq=29 name=ahci[0000:00:1f.2]
swapper 0 [000] 67144.520006: irq:irq_handler_exit: irq=29 ret=handledNAPI, or New API, was written to make processing packets of incoming cards more efficient. Hard interrupts are expensive because they cannot be interrupted. Even with interrupt coalescence (described later in more detail), the interrupt handler will monopolize a CPU core completely. The design of NAPI allows the driver to go into a polling mode instead of being hard-interrupted for every required packet receive. Under normal operation, an initial hard interrupt or IRQ is raised, followed by a SoftIRQ handler which polls the card using NAPI routines. The polling routine has a budget which determines the CPU time the code is allowed. This is required to prevent SoftIRQs from monopolizing the CPU. On completion, the kernel will exit the polling routine and re-arm, then the entire procedure will repeat itself.
NAPI 或新 API 的编写是为了更有效地处理传入卡的数据包。硬中断是昂贵的，因为它们不能被中断。即使有中断 合并（稍后详细描述），中断处理程序将独占一个 CPU 内核 完全地。NAPI 的设计允许驱动程序进入轮询模式而不是被 为每个需要的数据包接收硬中断。在正常操作下，会引发初始硬中断或 IRQ，然后是 SoftIRQ 处理程序 它使用 NAPI 例程轮询卡。轮询例程有一个预算，它决定了 允许代码的 CPU 时间。这是防止 SoftIRQ 独占 CPU 所必需的。完成后，内核将退出轮询例程并重新启动，然后整个过程将 重复自己。
If the SoftIRQs do not run for long enough, the rate of incoming data could exceed the kernel's capability to drain the buffer fast enough. As a result, the NIC buffers will overflow and traffic will be lost. Occasionally, it is necessary to increase the time that SoftIRQs are allowed to run on the CPU. This is known as the netdev_budget. The default value of the budget is 300. This will cause the SoftIRQ process to drain 300 messages from the NIC before getting off the CPU:
如果 SoftIRQ 运行时间不够长，传入数据的速率可能会超过内核的速率 足够快地耗尽缓冲区的能力。结果，NIC 缓冲区将溢出并且流量将 迷路了。有时，需要增加允许 SoftIRQ 运行的时间 中央处理器。这称为 netdev_budget。预算的默认值为 300。这将 导致 SoftIRQ 进程在离开 CPU 之前从 NIC 排出 300 条消息：
代码语言：javascript复制heidsoft@heidsoft-dev:~$ sysctl net.core.netdev_budget
net.core.netdev_budget = 300
heidsoft@heidsoft-dev:~$- http://arthurchiao.art/blog/linux-irq-softirq-zh/#1-什么是中断
- https://novoland.github.io/网络/2014/07/26/网卡中断负载均衡.html
- https://applezulab.netdpi.net/linux-prog/about-linux-smp_affinity
- https://web.archive.org/web/20200225050436/http://blog.yufeng.info/archives/2422
- https://github.com/coreos/bugs/issues/2135
- https://kernelbook.sourceforge.net/kernel-hacking.html/ioctls.html
- https://unix.stackexchange.com/questions/341947/what-is-the-difference-between-proc-interrupts-and-proc-softirq-in-linux
- https://people.kernel.org/dsahern/the-cpu-cost-of-networking-on-a-host
