Repository navigation
Conversation
…, and the runner records its own namespace PID
Found on the Pixel 9a on 2026-11-17, the first time Location reached a
fix on it (7.3.0-rc6-omamobile-push1-gbd32edee1132):
- The desktop's location setting. libgeoclue's GClueSimple goes through
xdg-desktop-portal's Location portal, which refuses every request while
org.gnome.system.location enabled is false, the default on a new
account ("org.freedesktop.portal.Error.NotAllowed: Location services
disabled"). Nothing turned it on, so with Location on, a fix in the
NMEA stream and the agent running, where-am-i and every libgeoclue app
got nothing. The agent unit now sets it true before it starts and false
after it stops, so it follows the switch (and the session: the next
login's start turns it on again).
- GeoClue and a restarted agent. GeoClue 2.8.2 refuses every client of a
user whose agent went away and registered again while the service kept
running ("Starting GeoClue client failed: ... AccessDenied: Geolocation
disabled for UID 1000"), and turning Location off and on restarts the
agent. Reproduced three times on the phone; a stop of geoclue.service
(it is D-Bus activated) made the next agent start work every time.
"omamobile-gnss on" and "off" now stop geoclue as root before the
caller's agent starts or stops, so the agent's registration goes to a
fresh service.
- The upgrade from before 0.2.11 stops geoclue and restarts the running
agents once (a restart, not start + try-restart, which would register
twice under the new service), so the setting is applied without a new
login.
- The namespace PID. omamobile-vendor-chroot wrote
awk '/^NSpid:/ {print $2}' /proc/self/status, which is awk's own,
short-lived PID, so "run" and "ns" failed with "nsenter: cannot open
/proc/N/ns/mnt" (the documented way to start gnssd by hand). It is now
read with a shell builtin, so /proc/self is the runner.
On the phone: Location off -> on -> off -> on with this CLI and agent
unit, each "on" answered at once (GeoIP first, then GPS GGA/RMC); without
it the second "on" was refused until geoclue restarted. The ns.pid fix
let "omamobile-vendor-chroot modem run /vendor/bin/hw/gnssd" enter the
running modem namespace.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Thank you for this, and for #862: three separate causes, each with its log line, is exactly what makes a report like that usable. I reproduced the Two things have to happen on my Pixel 7 Pro before I merge, and both are mine to do:
One suggestion for the tests: I'll report the 7 Pro's results here. |
|
One thing for when 0.2.11 is built for the repository, since this takes the version I asked for in #748: the published 0.2.10-1 has no I'll run this branch with #863 on my palmtop tomorrow and post the result here. |
|
Ran on my palmtop this morning with #863 (the kernel and versions are in my comment there): omamobile-gnss 0.2.11-1 from this branch, with the 9a's client from my local
One thing I couldn't check is the overlay's capture at the modem's stop: the NV overlay isn't turned on on my palmtop (no |
|
The clean reboot was this morning, so here is the modem's first stop with a correct Before the reboot, with the modem up:
At the stop ( That one line is all the journal has on the overlay for that boot. So the capture at the stop is reached now, and with the overlay off it keeps nothing. The unit's stop took 0.6 s. The boot after it came up as usual: the CP ONLINE with IMS voice and SMS and the data bearer, the stub bound read-only over I haven't turned the overlay on here, so a capture that keeps something is still untried on the 9a. |
….2.11" branches The stand-in vercmp only ever printed 0 or 1, so the branches that stop geoclue and restart the agent on an upgrade from before 0.2.11 never ran, and test/all only grepped for the two strings. The stand-in now answers -1, 0 and 1 like pacman's, and the hook runs as an upgrade to 0.2.11-1 from 0.2.8-1, 0.2.10-1 and 0.2.11-1: from before 0.2.11 geoclue stops before the agent restarts (by line order in the recorded calls); from 0.2.11 on geoclue is left alone and the agent is only started. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Results from my Pixel 9a (tegu, slot B, kernel 7.3.0-rc6-omamobile-push1-gbd32edee1132, omamobile-modem 0.1.26-1), and the test you suggested. The test. Description. Added under ns.pid: the fix also makes the NV overlay's capture at the modem's stop, and The upgrade. 0.2.11-1 built from this branch (with the client from The switch, done as the tile does it ( Your question: a modem stop with a correct That is the only overlay line. The overlay is off here, as on W4HO-ham's palmtop, so a capture that keeps something is still untried on the 9a. The next boot came up normally: modem and mobile data up, gnssd running, One thing I learned the hard way, for anyone testing this: on the 9a the modem can't be started a second time in the same boot ( |
|
Thank you both. @W4HO-ham, that run answers what I asked den-kloot about the 9a, and more: the upgrade from 0.2.10 with Location on, the switch four times with the setting following it and no refusal from GeoClue, Where this stands:
@den-kloot, one request from last time is still open: a sentence in the description that the |
|
Both cases again with After an off/on: a fix, just past 40 s. Location off and on as the tile does it, then So the first position from the receiver came about 46 s after the request, and my earlier 40 s run ended just before it. Until then the client got GeoClue's coarse positions (25 km). A second off/on two minutes later had the receiver's first position 9.7 s after the request, and 240 of them in the 120 s. So that case is "no fix yet", not a cause. The start is cold every time because gnssd gets no reference time or position: its NTP lookup fails inside the modem profile, and the extended ephemeris request finds no network: That isn't this PR's business, but it is why a first fix takes about 40 s here even with the sky in view. The first request after a boot: a fourth cause. I rebooted and ran gnssd started at 09:29:06, 20 s after ONLINE ( So a location request between the modem's start and gnssd's start leaves Location dead until the next boot or a toggle. The window is about the first minute after boot. The service exits at once there, GeoClue reconnects at once, and the start limit does the rest. The fix I'd suggest is in
I can send that as its own pull request on main once this one is in, as 0.2.12, unless you'd rather it went another way. Nothing in either run asked for a coordinate; the numbers above are counts, times and accuracies. |
derekross
left a comment
There was a problem hiding this comment.
@den-kloot, first an apology: your push and your results of 15:51 crossed my comment of 16:06, which asked again for two things you had just done. Both are in: the description now says the ns.pid fix makes the overlay's capture and nv-overlay diff|capture reach the profile, and the hook's test runs the "older than 0.2.11" branches. I ran that test and broke the hook three ways (no geoclue stop, a restart from 0.2.11 on, the threshold at 0.2.10); each one fails the assertion written for it. Thank you also for the real upgrade from 0.2.10-1 and the second clean stop with a correct ns.pid. Nothing more is needed from you here.
One note on the modem: a stop and start of the 9a's modem unit in one boot is on our "never" list (a second start once reset the palmtop silently). The guard refused it for you, which is what it is for; a reboot is the way, as you found.
@W4HO-ham, thank you, that settles the 40 s question: gnssd's own line puts the cold first fix 40.6 s after the session start, so that run ended just short, and it is "no fix yet".
The first request after a boot is a real fault, and it is on main, not in this pull request: omamobile-gnss-start exits at once while gnssd isn't up yet, GeoClue reconnects, the start limit trips and the NMEA socket stays failed for the boot. Yes please to a pull request of its own as 0.2.12 once this is in. Two things to keep in it: the wait goes before the hang marker is written, as the client check does, and a wait that gives up leaves no marker behind. Could you open an issue with your journal lines now, so it isn't buried here? I've seen #900 and will review it on its own.
What is left is mine: the two Pixel 7 Pro runs (the overlay's state and a stop there, and the Location tile off and on with the weather widget open), and rebuilding the client on the build machine before 0.2.11 is published. I'll post the runs here and merge after them.
…pt-in), with the runs on the published branch Location now works whenever the modem is up, and the modem's start at boot is opt-in for the reason the Modem cell names, so the cell takes the same form as the Modem and Suspend cells. The record has the runs on the published omamobile at 26867fc4e with derekross#864 and derekross#921: a GNSS fix in 25 of 28 boots from a request made before gnssd was up, each reboot made during a session, none skipping GNSS; GeoClue's own start stalled in the other 3, cause open; and five switch-offs and -ons with a fix 7.5 to 9.4 s after each request.
What and why
Parts 2-4 of #862, found on the Pixel 9a the first time Location reached a fix on it:
org.gnome.system.location enabledis false, the default on a new account (org.freedesktop.portal.Error.NotAllowed: Location services disabled). The GeoClue agent unit now sets it true before it starts and false after it stops, so it follows the switch.AccessDenied: Geolocation disabled for UID 1000), and the switch restarts the agent.omamobile-gnss on|offnow stopsgeoclue.service(D-Bus activated) before the caller's agent starts or stops, so the agent registers with a fresh service. The upgrade from before 0.2.11 does the same and restarts running agents once, so the setting is applied without a new login.omamobile-vendor-chrootrecorded awk's own PID (awk '/^NSpid:/ …' /proc/self/statusruns in a child), sorunandnsfailed withnsenter: cannot open /proc/N/ns/mnt. That is the documented way to start the 9a's gnssd by hand. The PID is now read with a shell builtin. The same file is what the modem profile's NV overlay uses, so the fix also makes the overlay's capture at the modem's stop, andomamobile-modem nv-overlay diffandcapture(and their timer), reach the running profile: with awk's dead PID they could never have worked, on any phone.Device and build
uname -r): 7.3.0-rc6-omamobile-push1-gbd32edee1132Checks (no phone needed)
tools/pr-checkpasses (againstc4623572, on the laptop)tests/all: passes excepttests/install-phone's "an interrupt on the other slot's screen", a signal-timing case in a file this PR doesn't touch. It failed on some runs and passed on others (including onmain) on a heavily loaded laptop.tests/packagesfor the package touched:packages/omamobile-gnss/test/allpassesversionbumped, and the new version is in the commit subjectCo-Authored-By:trailer, noSigned-off-by:On the phone
omamobile-vendor-chroot modem run /vendor/bin/hw/gnssdentered the running modem namespace and gnssd ran.userdata, weakens the storage guard, or reads registers from userspacetools/phone-smokereportphone-smoke: unknown
Kernel
7.3.0-rc6-omamobile-push1-gbd32edee1132, tools atunknown, 2026-11-17T11:25Z.Packages: hyprland 0.57.0-3.4, libcamera 0.7.2-4.25, mesa 1:26.2.4-1.4, omacam 0.3.3-1, omamessages 0.2.2-1, omamobile-apps 0.2.3-1, omamobile-audio 0.5.18-1, omamobile-device-tegu 0.2.18-1, omamobile-firstboot 0.1.5-1, omamobile-gnss 0.2.9-1, omamobile-keyring 20261026-1, omamobile-mirror 0.1.1-1, omamobile-mmd 0.1.4-1, omamobile-modem 0.1.21-1, omamobile-modem-sit 0.3.35-1, omamobile-omarchy 0.1.44-1, omamobile-owl-widgets 0.1.2-1, omamobile-sensors 0.2.1-1, omamobile-settings 0.1.2-1, omamobile-shell 0.2.127-1, omamobile-voice 0.1.0-1, omaphone 0.2.3-1, omarchy 4.5.0-3, omatel-agent 0.2.4-2, quickshell 0.3.2.r7.g2b4501a-1
(
omamobile-bt-addr@hci0failed on the previous boot with the old kernel too: unrelated.)🤖 Generated with Claude Code