I can confirm the overrun from comment #1 with real numbers. On a Remote
Login virtual monitor at 2560x1600 and scale 1.5,
clutter_stage_paint_to_buffer() reads back 2561x1601 into a 2560x1600
buffer (layout 1707x1067). I logged this from
meta_screen_cast_virtual_stream_src_record_to_buffer():
Virtual screen cast: view 1707x1067 @ 1.500 does not match buffer
2560x1600 stride 10240
On NVIDIA the same overrun crashes inside libnvidia-eglcore every time
(bug #2162030). On Mesa it shows up as this crash in
convert_ubyte_rgba_to_bgra().
Upstream, it is fixed on main (and in 51) by commit 93a178eaf, which
paints into a framebuffer of the negotiated stream size, so the odd-
height texture question from comment #3 doesn't come up. A reviewed
gnome-50 backport is pending:
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/5251 (issue:
https://gitlab.gnome.org/GNOME/mutter/-/work_items/5151). It applies
cleanly to Ubuntu's 50.1, and it fixes the NVIDIA crash for me in the
greeter and the user session. It would be great to have it in 26.04 if
upstream doesn't release it in a 50.x soon.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2156443
Title:
gnome-shell crashed with SIGSEGV in convert_ubyte_rgba_to_bgra() from
read_rgba_pixels() from _mesa_readpixels() from st_ReadPixels() from
read_pixels()
To manage notifications about this bug go to:
https://bugs.launchpad.net/mutter/+bug/2156443/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
