Skip to content
Closed
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
32 changes: 32 additions & 0 deletions docs/review/LESSONS.md
Original file line number Diff line number Diff line change
Expand Up @@ -1792,6 +1792,38 @@ never change, so reviews can cite them.

## Claims and evidence

- **L-344. A console quoted as "just before" a failure is checked against
that console's own timestamps.** I cited ACPM channel timeouts as
preceding the 9a's UFS resume failure in #152; that console has no ACPM
timeout at all, and the 25 s before its failing suspend hold only PAM and
audit lines. The timeouts were from a different boot of the same day.
*Check:* grep the file you are citing for the line, and compare the
bracketed uptimes, before writing "before" or "then" in a PR. *From:*
kernel #152, written and corrected the same night.
- **L-345. "The Nth of a series failed" is counted from the console, not
from the session you remember.** The 9a's UFS failure was called a late
sleep in a long boot; the console shows nine suspend entries and the
third one failing, which retired the sleep-count theory the record had
just acquired. *Check:* count the `PM: suspend entry` lines before the
failure. *From:* #166, round 1.
- **L-346. A flaky check is read for what it asserts before it is blamed on
the machine.** `test/owl-idle`'s paused-tick check failed on four PRs that
don't touch the home plugin, and under parallel load locally, which looked
like a slow runner. `tickDelay()` returns the time to the next boundary,
so the assertion is false for one second in every ten by construction,
and the CI message's own figure (554 ms) placed the clock 9451 ms into a
10000 ms period. *Check:* work the printed value back through the function
before calling a check timing-sensitive; 30 local runs that pass prove
nothing against a 10 % window. *From:* #207.
- **L-347. "A shared static means the instances collide" is checked against
how its index is assigned.** `eint_wake_mask_values[]` in
`pinctrl-exynos.c` is file-scope and written by three wake-capable
controllers on the 9a, which reads like a collision; `eint_num` is also
file-scope and never reset, so the numbering is continuous and the shared
array is the point. *Check:* find where the index comes from before
reporting the sharing. *From:* the 9a's sleep hang, 2026-10-30, a
candidate dropped before it reached a PR.

- **L-343. A `-dirty` release that stays the same across a fix doesn't
identify the tested build.** The 9a's NFC runs without the fix (10-28) and
with it (10-29) report the same `…-dirty` release, and #137 quoted that
Expand Down