From cab22e5a4ad5a24ca8c2fe4cf43257d86ac4257c Mon Sep 17 00:00:00 2001
From: vasilito <adminpupkin@gmail.com>
Date: Wed, 5 Aug 2026 14:22:28 +0300
Subject: [PATCH] docs: append REGRESSION + REMEDY record for task-8 ps2d
 read-to-string regression

Root cause: std::fs::read_to_string loops forever on kernel scheme
/scheme/serio/present (serio kread for HandleKind::Present never
produces EOF). Fix: bounded single-byte read in ps2d main.rs
(commit aeb2df8e on submodule/base branch). 14/14 host tests green.
QEMU gate expected: ps2d startup log BEFORE notify-wait ticks,
no kill line.
---
 .../task-8-ryzen-7000-x670e-compat.txt        | 58 +++++++++++++++++++
 1 file changed, 58 insertions(+)

diff --git a/.omo/evidence/task-8-ryzen-7000-x670e-compat.txt b/.omo/evidence/task-8-ryzen-7000-x670e-compat.txt
index 648a2ba4e7..634cdbf1d7 100644
--- a/.omo/evidence/task-8-ryzen-7000-x670e-compat.txt
+++ b/.omo/evidence/task-8-ryzen-7000-x670e-compat.txt
@@ -92,3 +92,61 @@ Single command (no build needed):
 - No `src/scheme/irq.rs` cpu- path handling touched
 - No retries/warning noise on absence
 - No recipe files modified
+
+---
+## REGRESSION + REMEDY RECORD (2026-08-05, QEMU verification)
+
+### Root cause
+
+`std::fs::read_to_string("/scheme/serio/present")` in
+`local/sources/base/drivers/input/ps2d/src/main.rs:38` loops forever on
+the i8042-present path. The kernel's `SerioScheme::kread` for
+`HandleKind::Present` (kernel `src/scheme/serio.rs:146-148`) always
+returns the verdict bytes (`b"1\n"` or `b"0\n"`) on every read without
+ever producing a zero-byte EOF. `read_to_string` in Rust std reads
+until `read()` returns 0, so it enters an infinite read loop — growing
+the accumulator string by 3 bytes per iteration — and never returns to
+caller code. ps2d blocks in `read_to_string` before any log output or
+`daemon.ready()` can execute, so init's 10s notify-wait timeout kills it
+with zero diagnostic output ("'ps2d' did not notify readiness within 10s;
+killing and skipping").
+
+The absent path (kernel verdict "0\n") is equally affected but was not
+caught because the QEMU configs used for absent-path testing happened to
+not exercise `read_to_string` (or the absent-path tests were host-only
+verdict logic tests, not QEMU boots). The host test suite (14/14 GREEN)
+tests the verdict DECISION logic in isolation — it does not exercise the
+actual `read_to_string` syscall path on a running Redox kernel.
+
+### Remedy
+
+Commit `aeb2df8e` on `submodule/base`:
+
+- **`drivers/input/ps2d/src/main.rs`** — Replaced `std::fs::read_to_string`
+  with a bounded single-byte read (`read_serio_present()`): opens
+  `/scheme/serio/present`, reads at most 2 bytes into a stack buffer,
+  returns the first byte. Only the first byte matters (b'1' = present,
+  b'0' = absent); the '\n' is discarded.
+
+- Added a `log::info!` on the present path BEFORE `InputProducer` opens
+  ("ps2d: PS/2 controller present; starting serio device handshake") so
+  future boot logs show ps2d's progress through the startup sequence
+  before any scheme I/O blocking point.
+
+Host tests: 14/14 GREEN (`cargo test --manifest-path
+local/sources/kernel/host-tests/i8042-verdict/Cargo.toml`). All verdict
+decision-table tests pass unchanged — the fix only changes the I/O
+mechanism, not the decision logic.
+
+### Expected QEMU gate observable
+
+On the i8042-present path, the orchestrator's QEMU gate should see:
+- **BEFORE** the first `notify-wait 'ps2d' tick` (within ~50ms of
+  "Started PS/2 driver"): the line
+  `ps2d: PS/2 controller present; starting serio device handshake`,
+  followed by `ps2d: registered producer handle, listening on serio/0
+  (keyboard) and serio/1 (mouse)`.
+- **ABSENT**: the kill line `'ps2d' did not notify readiness within 10s;
+  killing and skipping`.
+- Keyboard input on `/scheme/serio/0` works in QEMU after ps2d's
+  successful notify.