Comment # 15 on bug 1137067 from Ross Green

I suspect this is the same problem.

The trick to finding out for sure is have a look at the stack trace.

Even though the nouveau driver baffs, with a little bit of luck you might get
enough written to the /var/log/boot.msg that will include the stack trace of
the Null pointer. If you look for the nouveau establishment announcments,
you might see the lines:

Call Trace:
<4>[    5.990122]  ? drm_dp_dpcd_access+0x69/0xf0 [drm_kms_helper]
<4>[    5.990129]  ? drm_dp_dpcd_read+0x2e/0xc0 [drm_kms_helper]
<4>[    5.990137]  ? drm_dp_cec_set_edid+0x4e/0x140 [drm_kms_helper]
<4>[    5.990183]  ? nouveau_connector_detect+0x1fd/0x440 [nouveau]
<4>[    5.990220]  ? nouveau_connector_detect_lvds+0x132/0x180 [nouveau]
<4>[    5.990254]  ? nouveau_connector_detect_lvds+0x132/0x180 [nouveau]
<4>[    5.990262]  ? drm_helper_probe_single_connector_modes+0xc3/0x6b0
[drm_kms_helper]
<4>[    5.990270]  ? __alloc_pages_nodemask+0xda/0x230
<4>[    5.990278]  ? drm_setup_crtcs+0x139/0xa80 [drm_kms_helper]
<4>[    5.990285]  ? __drm_fb_helper_initial_config_and_unlock+0x2d/0x400
[drm_kms_helper]
<4>[    5.990293]  ? mutex_lock+0xe/0x30
<4>[    5.990327]  ? nouveau_fbcon_init+0x14f/0x1c0 [nouveau]
<4>[    5.990361]  ? nouveau_drm_device_init+0x184/0x7f0 [nouveau]
<4>[    5.990396]  ? nouveau_drm_probe+0x1eb/0x2b0 [nouveau]


this is the stack unwind of the calls leading up to the problem.


there are several calls from within drm_kms_helper.
plus nouveau_connector_dector calls.

This would indicate kernel modeset is involved, with spurious data.

Indeed if you set nomodeset you can boot the kernel, but you end up with a
screen that is usually in a VGA resolution and not the real functionality you
want from the system.

Paul if you are happy to undertake a bisect to narrow down the offending
problem code by all means do.

I suspect part of the problem might be using some of the upstream nouveau
driver code without all the required extras.

I now have several systems that exhibit the above problem.
Unfortunately from a work perspective NVIDIA drivers are used to allow use with
various bits of CUDA hardware and software.

What I find very frustrating with this bug is that it can affect standard
installs, where the newer driver gets downloaded and the resulting system ends
with just a blank screen and no other visible information to tell what is going
on.

I spent several hours the other day looking at Linus' tree following all the
updates to the nouveau driver. It might be worth trying one of the more recent
kernels with nouveau driver and see if the problem does exist in the upstream
code.

Ross