gps_power_control() mixed the two GPIO conventions on the same pins. It
configured lines with GPIO_OUTPUT_HIGH / GPIO_OUTPUT_LOW, which Zephyr defines
as physical levels, then drove the same lines with gpio_pin_set_dt(), which is
logical. On a line declared GPIO_ACTIVE_LOW those are opposite, so the first
call asserted the reverse of every call after it.
Two lines are actually broken by this.
gps-enable, on the boards declaring it active-low. Most of them reach the block
named for the T1000-E rather than the "simple boards" path, because
HAS_T1000_GPS_CONTROL is (HAS_GPS_VRTC || HAS_GPS_RESET || HAS_GPS_SLEEP) and a
bare gps-reset alias is enough to route a board there. That block drives the
enable physical HIGH unconditionally on every power-on, and power-off lands on
physical HIGH too, so the line is de-asserted in both states and the module is
never powered at all.
gps-reset, on the six boards declaring it active-low: heltec_t096, heltec_t1,
heltec_t114, thinknode_m1, heltec_wifi_lora32_v4 and _v43. The documented
power-on sequence is "assert reset, hold 10 ms, release", written as
GPIO_OUTPUT_HIGH then gpio_pin_set_dt(0). On an active-low line both of those
are physical HIGH, so the pulse never asserts and the module starts
uncontrolled -- the condition the power-off comment says this code prevents.
heltec_wifi_lora32_v4's overlay documented the no-op rather than fixing it;
that comment is updated here, with _v43's copy.
vrtc, sleep, rtcint and resetb are converted too. Every board declaring them is
active-high, so those change nothing today; they are converted so the root
cause does not survive in four more places.
Converting is safe by construction. gpio_pin_configure() flips the init level
only when both GPIO_OUTPUT_INIT_LOGICAL and GPIO_ACTIVE_LOW are set, then
strips INIT_LOGICAL, so on an active-high pin GPIO_OUTPUT_ACTIVE produces a
flags word byte-identical to GPIO_OUTPUT_HIGH. Behaviour changes only where a
line is declared active-low, which is only where it is currently wrong.
gps_power_off_for_shutdown() had the same inversion and it mattered most there:
on the active-low boards the old physical LOW asserted the enable, powering the
GPS on as the device entered System OFF rather than powering it down. Its
header comment also described a narrower scope than the function has, naming
only the power enable where it configures vrtc, reset, sleep, rtcint and
resetb as well; that is corrected here.
Comments that described a level this change now drives logically are reworded
to asserted/de-asserted. Comments stating genuine electrical facts, such as a
pin floating LOW without its pull-up, are left as they are, since rewording
those would make them wrong.
Verified on hardware on a Heltec V4-R8, whose gps-enable is active-low: with
this change the boot log reports "GPS power GPIO configured, set ACTIVE" and
the L76K streams NMEA (210 sentences in 35 s), where without it the log reports
"set HIGH" and the module never powers on. Also built for heltec_wifi_lora32_v4
and _v43 (active-low enable and reset), thinknode_m1 and heltec_t114
(active-low reset) and t1000_e (active-high throughout).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The V4-R8 is an ESP32-S3R8 with 8 MB octal PSRAM, where V4 and V4.3 are S3R2
with 2 MB quad. Octal PSRAM claims GPIO33-37, which moves Vext 36->40, the LED
35->46 and VGNSS_Ctrl 34->42, and drops ADC_Ctrl and GNSS_RST. CTX binds as
lna-bypass-gpios rather than tx-enable-gpios, so the board is a fourth one
where set radio.fem.rxgain works; the CLI reference and the register-patch help
text say so.
The allowlist change is needed for the board to be added correctly. CMake
MATCHES is a substring search, so the existing heltec_wifi_lora32_v4 entry also
matches heltec_wifi_lora32_v4_r8 and the new board would get CONFIG_PM=y
without anyone deciding it should, on a path pm_esp32.conf marks as not yet
validated on hardware. Anchoring preserves the five boards that were intended
to match, including heltec_wireless_tracker_v2, which previously qualified only
by substring and is now listed explicitly.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
main_repeater.cpp and main_room_server.cpp applied the GPS duty interval
~55 lines before data_store.loadPrefs() ran, so getNodePrefs()->gps_interval
still held the initNodePrefs() value of 300 s.
That 300 is the *companion* default. loadPrefs() is not a pure read on these
roles: its first-boot path is the only place the 48 h repeater/room-server
default is established, and it persists that to flash so later code can treat
flash as authoritative. A stored value is likewise only in the struct after
loadPrefs() returns. Reading gps_interval before that call therefore cannot
see either the role default or the operator's setting -- only the companion
constant.
Two consequences, on every boot:
- A stock GPS-equipped repeater duty-cycled its GPS every 5 minutes instead
of every 48 hours, ~576x more wakeups than intended. On battery nodes this
is a straight power regression.
- Any operator value never reached the hardware. "set gps duty 0" (always-on)
persisted to flash correctly, and `get gps duty` reports the manager's
now-effective value, so it answered 300 -- readable as a stale default
rather than as the setting being ignored.
The GPS manager itself is correct: gps_start_acquiring(), the fix-done
callback, gps_enable() and the fresh-fix request all guard standby on
gps_duty_cycling(), so the always-on invariant holds once the manager is
actually told the interval is zero.
Same ordering class as the two bugs already documented in these files -- radio
frequency binding and the LED master switch -- where CLI readback looked right
and the hardware disagreed. main_companion.cpp has always had the correct
order (loadPrefs() at 1461, gps_set_poll_interval_sec() at 1616) and is
unaffected.
Move the interval and the mode switch that starts acquisition to just after
loadPrefs(). Callback registration stays where it was: it has no prefs
dependency and must precede acquisition.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Both emit ENTER via a zephyr,input-longpress filter (single button),
same category the Kconfig help text already documents as needing
CONFIG_ZEPHCORE_UI_CONFIRM_WINDOW_MS=3000 instead of the 500ms
default -- two ~1s holds don't fit in 500ms, so the second long press
on shutdown/DFU/offgrid always arrives after the window closed and
just re-arms instead of confirming. T096 already had this set
(board.conf:38); WT2 and V4.3 didn't, apparently missed when their
board.conf was written.
Preconfigured VS Code dev container with the Zephyr SDK, toolchains, west
and vendor blobs matched to CI, so a build can be reproduced without
installing prerequisites on the host.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
liquidraver