https://bugs.kde.org/show_bug.cgi?id=522307

zosima <[email protected]> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[email protected]

--- Comment #11 from zosima <[email protected]> ---
Confirming this on a second distro, with three additional data points: the
implicated pam_kwallet6 version matches, suspend/resume is NOT required to
trigger it, and downgrading pam_kwallet6 (your bisect) eliminates the problem.

I have posted full forensics in bug 522615, which appears to be the memory-leak
face of this same loop — in my case the end state was far more severe than
sustained CPU: kwalletd6 leaked to ~53 GB RSS at ~1 GB/min and was OOM-killed
by the kernel.

SOFTWARE VERSIONS
OS: Fedora 44 (KDE spin), kernel 7.0.14-201.fc44.x86_64
Plasma: 6.7.1 (Wayland)
kf6-kwallet / kwalletd6: 6.27.0-1.fc44
pam-kwallet (pam_kwallet6): 6.7.1-1.fc44  <-- the version your snapshot bisect
implicates
Secret Service backend: ksecretd

OBSERVED
- Fresh boot + normal PAM login. journalctl confirms zero suspend/resume events
in the boot ("PM: suspend" count = 0) — so the trigger here was the login
handoff itself, not the screenlocker path.
- ksecretd (started via pam: /usr/bin/ksecretd --pam-login 13 14) began
consuming sustained CPU from login: ~32% of a core, 21m57s CPU time over 68
minutes of wall clock.
- kwalletd6 was D-Bus-activated 8 minutes after login and its RSS grew
continuously until the kernel OOM-killed it at ~55 GB anon RSS (details in bug
522615).
- The instant kwalletd6 died, ksecretd CPU dropped to zero (measured 0
scheduler ticks over 5 s) — consistent with your dbus-monitor finding that
kwalletd6 drives the loop, and with "killall kwalletd6 stops it".
- Both daemons logged portal registration failures at startup:
    kwalletd6: Failed to register with host portal QDBusError(..., "Could not
register app ID: App info not found for 'org.kde.kwalletd'")
    ksecretd: "... Connection already associated with an application ID"
- The respawned kwalletd6 immediately logged Secret Service *client* errors:
    received unexpected result type ao from Completed signal instead of
expected o
    GTask secret_service_real_prompt_async (...) finalized without ever
returning
  i.e. kwalletd6 acting as a libsecret client of ksecretd, matching the
suspected SecretServiceClient::defaultCollection() path in comment 4.

CONFIRMED: DOWNGRADE FIXES IT
Fedora never shipped pam_kwallet6 6.7.0, so I downgraded to 6.6.4-1.fc44
(pre-regression). After reboot: wallet unlocks normally via PAM, ksecretd
consumes zero CPU (0 scheduler ticks over 10 s, 0:00 accumulated after several
minutes — previously ~32% from the first minute), kwalletd6 RSS flat at ~60 MB.
This independently confirms your bisect on a second distribution, and extends
the trigger from "suspend/resume + unlock" to the initial PAM login handoff as
well.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to