cpuset: count loadavg tasks inside the cpuset · Pull Request !1885 · OpenCloudOS/OpenCloudOS-Kernel

**Problem**

The `running/total` pair printed by the cpuset loadavg view mixes scopes: the numerator counts tasks of the selected cpuset (`cpuset_nr_running()` iterates `css_task_iter`), but the denominator is the **global** `nr_threads` counter. Processes outside the cpuset move the reported total of an otherwise untouched cgroup, so the ratio cannot be used to reason about the cgroup it is read from.

**Root cause**

The show helper copied the `/proc/loadavg` format, where `nr_running/nr_threads` is a system-wide pair, but kept only the per-cpuset semantics for the running count.

**Change**

Count the total during the same task iteration that counts the running tasks and print that cpuset-local number as the denominator.

**Verification**

Two-vCPU QEMU guest, cgroup v2 pod (`cpuset.cpus 0-1`) holding exactly one sleeping task:

- unfixed kernel: `cat pod/cpuset.loadavg` reports `0/63`, and `0/83` after twenty `sleep` processes are started *outside* the pod (pod membership unchanged, 1 task);
- fixed kernel: both reads report `0/1` while the same twenty outside sleepers run (the last-pid field still grows, proving they exist);
- regression: the load averages themselves and the last-pid field are unchanged.

`scripts/checkpatch.pl --strict`: 0 errors, 0 warnings.

类型	指派人员	状态
审查	zoucao Huang Cun Jianping Liu	已完成 (0/0人)
测试	Jianping Liu	已完成 (0/0人)

GVPOpenCloudOS / OpenCloudOS-Kernel

搜索帮助

GVP OpenCloudOS / OpenCloudOS-Kernel