Bug #1160: fmd assertion fails on boot on Dell C6145 - illumos gate - illumos

Actions

Copy link

Bug #1160

open

fmd assertion fails on boot on Dell C6145

Added by Rich Ercolani about 11 years ago. Updated almost 7 years ago.

Status:

New

Priority:

Normal

Assignee:

Category:

Start date:

2011-06-27

Due date:

% Done:

Estimated time:

Difficulty:

Medium

Tags:

needs-triage

Gerrit CR:

Description

On fresh boot on a Dell C6145 (in this case, a 4x12-core Opteron box), fmd dies on boot with "Assertion failed: comma != NULL, file ../../common/pcibus/did_props.c, line 401, function dev_for_hostbridge"

I suspect that http://opensolaris.org/jive/thread.jspa?threadID=117229 is the same bug - since the C6145 has an external PCIe port for connecting to e.g. a Dell C410x external PCIe chassis.

oi_148, will attach any dumps residing on the machine, as well as prtconf and similar, as soon as I bring up networking on the machine.

Files

Download all files

ls_devices.gz (4.06 KB) ls_devices.gz	ls -lR /devices output	Rich Ercolani, 2011-06-27 09:45 PM
prtconf.gz (538 Bytes) prtconf.gz	prtconf output	Rich Ercolani, 2011-06-27 09:45 PM
prtconf_p.gz (351 Bytes) prtconf_p.gz	prtconf -p output	Rich Ercolani, 2011-06-27 09:45 PM

Actions

Copy link

Updated by Rich Lowe about 11 years ago

In the mentioned thread, Mike mentions the range between his laptop working and not working was onnv_123 to onnv_126, within that range falls:

changeset:   3986:8f772ca46fa3
user:        myers
date:        Fri Apr 06 15:57:19 2007 -0700
description:
        6472670 Internal pci bus numbering should be independent of bios enumera
tion order (for non-x8400 systems)

Which changed PCI enumeration and made it persist. Some of the code explicitly states that it will now discover childless root-bridges which would previously missed. Seems suspiciously fitting.

I don't know much about this area, but I would suspect that the right thing to be doing is making fmd more robust.

Actions

Copy link

Updated by Rich Ercolani about 11 years ago

File ls_devices.gz ls_devices.gz added
File prtconf.gz prtconf.gz added
File prtconf_p.gz prtconf_p.gz added

Core dumps can be found at:
http://skysrv.pha.jhu.edu/~rercola/bitbucket2/

As well as copies of the attached outputs

Actions

Copy link

Updated by Rich Lowe about 11 years ago

Project changed from OpenIndiana Distribution to illumos gate

Actions

Copy link

Updated by Rich Lowe about 11 years ago

As the fm-discuss thread suggests, it looks like we assume any node which
pci_hostbridges_find discovers also have a driver attached
(by way of DEVprop_set, which pulls their devfs path and lets dev_for_hostbridge attempt to adjust it).

It looks like the path we get back from di_devfs_path is "/pci", because we have no parent nodes and di_bus_addr() is NULL, which does, I think, point toward no driver being attached as the linked thread states.

I think this means that fmd needs to be more resilient, as above, rather than enumeration being incorrect.

Actions

Copy link