This might well be a duplicate of #770130 and #776911 (but not #775235,
IMO) but I'm filing it as its own bug because you might find it useful
to have a separate bug# for something that at least one Debian developer
can definitely reproduce. I can test new versions of stuff on the
affected hardware if necessary, with a bit of delay while I borrow
peripherals from other machines.

Filing this as "normal" severity because, while it does make GNOME
unusable on the affected system, the hardware is a decade old. Please
adjust as appropriate.

Summary of hardware: an old Dell desktop with Pentium 4 CPU and i865g
on-board graphics. I wouldn't normally bother with such old hardware,
but it's the oldest x86 I have, and I was trying to find something with
as few CPU extensions as possible to disentangle whether #770130,
#776911 and #775235 are really the same bug.

Summary of software: a fresh jessie installation with the standard,
GNOME and ssh-server tasks, upgraded to today's versions.

Steps to reproduce:

* install
* boot the machine (to the gdm prompt, which is actually GNOME Shell)
* log in (also to GNOME Shell)

Expected result:

* the user's GNOME Shell works just as well as the gdm greeter did

Actual result:

* See logs below for full details, but the short version is:
* [drm] GPU HANG: ecode -1:0x00000000, reason:
  Command parser error, iir 0x00008000, action: continue
  (twice)
* GNOME Shell subsequently crashes with an assertion failure in Cogl:
  May 17 01:20:15 debian gnome-session[952]: (gnome-shell:1092):
  Cogl-ERROR **: Failed to create texture 2d due to size/format
  constraints
* gnome-session tries to restart GNOME Shell a couple of times,
  and Shell crashes in the same way
* gnome-session gives up and displays the "fail whale" ("Oh no!
  something has gone wrong" etc.)

Steps that *do not* reproduce this bug:

* Restarting gdm 5 times without logging in
  (in other words, whatever the user's gnome-shell is doing to trigger
  this, Debian-gdm's gnome-shell doesn't do it)

Things that do not affect this bug, which lead me to believe that it is
not the same issue as #775235:

* Replacing mesa with a version recompiled in current jessie
  (just to rule out any side-effect that it might have had for the
  other rebuilds)
* Replacing mesa with a version built with llvm3.4
  as per <https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=775235#15>
* Replacing mesa with a version with the rejected upstream patch
  for CPU feature detection
  as per <https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=775235#32>

Logs below include everything from
`reportbug --template xserver-xorg-video-intel` on the affected machine,
the output of `sudo journalctl` on the affected machine, and the
crash dump from /sys/class/drm/card0/error. Irrelevant IP
addresses/hostnames
replaced by CENSORED.

The log datestamps are way off because the machine's CMOS battery ran
out since I last used it (told you this was old hardware!), but
hopefully that doesn't affect anything.

Regards,
    S