https://bugs.kde.org/show_bug.cgi?id=516128
Kanishk Ranjan <[email protected]> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |[email protected] --- Comment #2 from Kanishk Ranjan <[email protected]> --- I can’t reproduce this on a similar stack. Environment: Ubuntu 26.04.1 LTS, Plasma 6.6.6 (KWin Wayland), Frameworks 6.24.0, Falkon 25.12.3, libqt6webenginecore6 6.10.2 (same QtWebEngine as the report), software rendering (llvmpipe). While running several Gemini queries, CPU spikes briefly while a response streams, then returns to idle. Real fd counts from /proc/PID/fd stay flat: main falkon process ~130, renderer ~32, zygotes ~13. Note that sudo lsof | grep falkon | wc -l reports [N] on the same session, because lsof prints a row per thread and also lists memory-mapped files. The ~24k figure in the report probably overstates the real fd count by a large factor. Could you, during the hang: Run ls /proc/$(pgrep -o falkon)/fd | wc -l a few times to see whether the real count grows. Run ls -l /proc/$(pgrep -o falkon)/fd | awk '{print $NF}' | sed 's/:.*//' | sort | uniq -c | sort -rn | head to show which fd types dominate. Say which GPU/driver you’re using on sway. My test used software rendering, so if this is a GPU/dmabuf path issue on wlroots compositors, I wouldn’t hit it. -- You are receiving this mail because: You are watching all bug changes.
