Claude review: cgroup/dmem: accept a single region when writing to attributes

[Claude Code Review Bot <claude-review@example.com>]

Patch Review

**Verdict: Good, minor observations**

This replaces the `while` loop that iterated over newline-separated regions with a flat single-region parse. The motivation (lack of atomicity when multiple regions could partially succeed) is valid.

The control flow is clean. One observation:

- After `strstrip(buf)` at line 737, the `strsep(&buf, " \t")` call at line 741 will always return a non-empty string if `buf[0]` was non-zero (checked at line 738), so the `!region_name[0]` check at line 742 is technically redundant but harmless as defensive coding.

- The `out_put` label correctly drops the region ref via `kref_put`. The early returns before the region lookup don't need ref cleanup, and the error path after `get_cg_pool_unlocked` correctly jumps to `out_put`. The refcounting is correct.

- **Minor nit**: The patch doesn't strip a trailing newline before parsing. If userspace writes `"region_name 1024\n"`, `strstrip()` at the top handles that. But if input is `"region_name 1024\n\n"` or `"region_name 1024\nother_region 2048\n"`, the `memparse` in `dmemcg_parse_limit` would hit the `\n` and return `-EINVAL` via `*end != '\0'`. This is the desired behavior (rejecting multi-region input), but it would be slightly more user-friendly to strip trailing whitespace from `buf` after `skip_spaces`. In practice `strstrip` already strips trailing whitespace from the entire buffer, so this is fine — `buf` after `strsep` points past the separator, and `skip_spaces` handles leading whitespace, while `strstrip` already trimmed trailing whitespace from the whole original buffer. No issue here.

---
Generated by Claude Code Patch Reviewer