|WARNING| [RFC] eal: allow setting random number generator seed

[dpdklab at iol.unh.edu]

Test-Label: ai-code-review
Test-Status: WARNING
https://dpdk.org/patch/169429

_AI Code Review_

=== Patch Review: series.patch (via Claude) ===
Review date: 2026-09-11

# DPDK Patch Review

## Summary
This patch adds the ability to override the automatic random seed via the `DPDK_RANDOM_SEED` environment variable. Overall, the implementation is straightforward, but there are several issues that need to be addressed.

---

## Errors

### 1. Missing documentation (Error)
The `DPDK_RANDOM_SEED` environment variable is undocumented. All environment variables must be documented in the EAL programmer's guide and release notes.

**Fix:**
- Add documentation to `doc/guides/prog_guide/env_abstraction_layer.rst` describing the variable, its format, and its purpose
- Add a release notes entry in `doc/guides/rel_notes/release_26_03.rst` (or current release)

### 2. `getenv()` in library code (Error)
The guidelines forbid `getenv()` in `lib/` directories. See Forbidden Tokens section.

**Rationale:** Environment variable access creates hidden dependencies and makes behavior non-deterministic across deployments.

**Fix:**
Add a new EAL command-line option instead:
```c
--random-seed <value>
```
This makes the configuration explicit and visible in application invocation, and follows DPDK's standard pattern for runtime configuration.

---

## Warnings

### 1. Missing test coverage (Warning)
New functionality should have test coverage in `app/test`. There should be a test that:
- Sets the seed via the new mechanism
- Verifies that `rte_rand()` produces the expected deterministic sequence
- Verifies that different seeds produce different sequences
- Verifies that an invalid seed format returns an error

**Suggested test:**
```c
static int
test_rand_seed_override(void)
{
	uint64_t seed = 0x123456789ABCDEF0ULL;
	uint64_t val1, val2;

	/* Set seed and get first value */
	rte_srand(seed);
	val1 = rte_rand();

	/* Reset with same seed, should get same sequence */
	rte_srand(seed);
	val2 = rte_rand();

	TEST_ASSERT_EQUAL(val1, val2, "Same seed should produce same sequence");

	return TEST_SUCCESS;
}
```

### 2. No validation of seed value range (Warning)
`strtoull()` accepts any value from 0 to `UINT64_MAX`, but seeds of 0 may produce degenerate RNG behavior depending on the algorithm. Consider whether seed=0 should be rejected or documented as potentially producing low-quality randomness.

### 3. Error message has trailing newline (Warning)
The `rte_exit()` format string includes `\n`, but `rte_exit()` already adds a newline.

**Fix:**
```c
rte_exit(EXIT_FAILURE,
	 "invalid DPDK_RANDOM_SEED: %s", env);
```

---

## Style Issues

### 4. Implicit comparison (Info)
The condition `*env != '\0'` follows DPDK style (explicit comparison). However, `env != NULL` could optionally use `likely()` if this is a hot path during initialization (though it's only called once at startup, so this is purely informational).

Current code is acceptable per guidelines.

---

## Additional Observations

### Variable scope
The variable `env` could be declared at point of use (C99 style):
```c
const char *env = getenv("DPDK_RANDOM_SEED");
```
This is already done correctly.

### errno handling
The errno check is correct: `strtoull()` sets `errno` to `ERANGE` on overflow, and checking `errno != 0` catches this. The subsequent check `*end != '\0'` catches invalid characters.

---

## Recommendation

The patch needs two changes before acceptance:

1. **Replace `getenv()` with an EAL command-line option** (Error) - see Forbidden Tokens
2. **Add documentation** (Error) - both programmer's guide and release notes

The test coverage (Warning) should be added for completeness, though lack of tests is not a blocker for an RFC.

After addressing the errors, this is a useful feature for reproducible performance testing.