Bug 198573

This was observed by SGI QA on SGI Altix as well.

Starting HAL daemon: hald(15205): unaligned access to 0x60000ffffe0675dc,
ip=0x200000000052a630
hald(15205): unaligned access to 0x60000ffffe0675dc, ip=0x200000000052a521

This also was with 2.6.16-1.2290_EL (the rhel5 alpha kernel).

David,  have you see anything like this?  It looks like HAL is trying to probe a
device and running into a bug either in the way HAL probes the device, the way
the kernel accesses it or in the device itself.

Eric and Brian can you attach the output of lshal?  Note that in order for this
to be fixed in RHEL4 you will need to go through your support rep. RHEL5 we
should be able to get in if we can identify the issue.

Reading the bugs again I noticed this was a RHEL5 only issue but was filed
against RHEL 4.  Please confirm.

Created attachment 133241 [details]
lshal output

WRT Comment #3, SGI is seeing this in RHEL% so if we convert this bug to a
RHEL4
bug, I can opena  RHEL5 one :)

With RHEL5 Alpha in mind... here is the requested lshal attachment.

Hi,

So, does hald crash? I guess not since you can get lshal output. 

Also, do you see any peculiar output running

 # /usr/sbin/hald --daemon=no --verbose=yes

as root (remember to stop the haldaemon service, 'service haldaemon stop').

Thanks,
David

John,

This is a rhel5 issue only.  My account didn't have access to file against rhel5
when I filed this bug, and now that it does I'm unable to change it.  I've been
working with Chris Williams at RH to try and get this fixed...  If you could
change it to "RHEL Beta" and version "5.0.0", that'd be great!

Also, hald doesn't crash.  I didn't see anything peculiar in the output of the
command you ask us to run, but I did see 'error', 'fail', and 'warn' show up a
few times:

[root@min ~]# /usr/sbin/hald --daemon=no --verbose=yes &>hald-log
hald(2931): unaligned access to 0x60000fffff24f51c, ip=0x200000000050a630
hald(2931): unaligned access to 0x60000fffff24f51c, ip=0x200000000050a630
hald(2931): unaligned access to 0x60000fffff24f51c, ip=0x200000000050a521
hald(2931): unaligned access to 0x60000fffff24f51c, ip=0x200000000050a521

[root@min ~]# grep -i error hald-log
14:12:55.771 [E] acpi.c:795: Couldn't open /proc/acpi/battery: Error opening
directory '/proc/acpi/battery': No such file or directory
14:12:55.771 [E] acpi.c:795: Couldn't open /proc/acpi/ac_adapter: Error opening
directory '/proc/acpi/ac_adapter': No such file or directory
14:12:55.771 [E] acpi.c:795: Couldn't open /proc/acpi/button/lid: Error opening
directory '/proc/acpi/button/lid': No such file or directory
2955: 14:12:56.478: probe-input.c:181: Error: EVIOCGID failed: Inappropriate
ioctl for device
[root@min ~]# grep -i fail hald-log
2936: 14:12:55.784: probe-smbios.c:140: Failed to execute dmidecode!
2955: 14:12:56.478: probe-input.c:181: Error: EVIOCGID failed: Inappropriate
ioctl for device
[root@min ~]# grep -i warn hald-log
2972: 14:13:00.038: probe-volume.c:592: warning: partition_number=2 not in [0;1[

Bryan

Hi Bryan,

is it possible you can attach the full log and also pipe stderror to stdout so
we can see where this

 hald(2931): unaligned access to 0x60000fffff24f51c, ip=0x200000000050a521

happens? Or maybe that message comes from the kernel and is just printed to the
console from the kernel.... Either way, having the full log would be useful. Thanks.

I will need to find a box to reproduce this.

Created attachment 133269 [details]
Requested log file

The unaligned access messages are definately coming from the kernel.  For the
most part they're harmless.  They basically tell you that the CPU had to use an
inefficient method for accessing the requested memory.	However, these messages
can be confusing to our customers, so we'd like them fixed before rhel5 ships.

Created attachment 136145 [details]
Fix hald unaligned access messages

Patch submitted by Yanmin Zhang to fix unaligned access messages.

I think we fixed this upstream some time ago. Please try with hal-0.5.8
available from here 

(you need to rebuild the SRPM for ia64, note that this src.rpm incorrectly
doesn't BR libvolume_id-devel (fixed in later version) so please install that
RPM separately. Thanks.)

 http://people.freedesktop.org/~david/hal-0.5.8.1-fc6/

Thanks. (built for FC6 but should work on RHEL5 too)

Should be fixed in 0.5.8.1-2 otherwise please reopen.

Yes, seems fixed - checked rhel5 rc snap3.