Claude review: tracepoint: Add trace_call__##name() API

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

Patch Review

**LOCKDEP assertion missing from `trace_call__##name()`**

In `__DECLARE_TRACE`, the existing `trace_##name()` includes a LOCKDEP RCU-watching assertion:

```c
if (IS_ENABLED(CONFIG_LOCKDEP) && (cond)) {
    WARN_ONCE(!rcu_is_watching(),
              "RCU not watching for tracepoint");
}
```

The new `trace_call__##name()` in this patch omits it entirely:

```c
static inline void trace_call__##name(proto)
{
    __do_trace_##name(args);
}
```

The cover letter claims "preserving all other correctness properties of the normal path (RCU-watching assertion...)" but the implementation does not actually do this. While one could argue the assertion already fired from the preceding `trace_foo_enabled()` call (which does *not* check RCU watching — it only checks the static branch key), the LOCKDEP check in `trace_##name()` runs unconditionally regardless of the static branch, specifically to catch RCU misuse even in testing where the tracepoint may not be enabled. Dropping it from `trace_call__##name()` changes debug-build behavior.

The same concern applies to `__DECLARE_TRACE_SYSCALL`, though there `might_fault()` *is* preserved, which is good.

Consider adding the LOCKDEP check to both `trace_call__` variants, or explicitly documenting in the commit message that it is intentionally omitted and why.

**Commit message inconsistency**: The commit message says:
```
trace_invoke_foo(args);   /* calls __do_trace_foo() directly */
```
This still uses the old `trace_invoke_` naming from v1, but the actual API is `trace_call__`. This should be updated to match.

**Code looks correct otherwise**: The three insertion points (regular, syscall, stub) are the right places, and the stub correctly compiles to an empty function.

---
Generated by Claude Code Patch Reviewer