Bug 83231 - install hangs on SMP system
Summary: install hangs on SMP system
Alias: None
Product: Red Hat Linux
Classification: Retired
Component: kernel
Version: 9
Hardware: All
OS: Linux
Target Milestone: ---
Assignee: Arjan van de Ven
QA Contact: Brian Brock
Depends On:
TreeView+ depends on / blocked
Reported: 2003-01-31 16:26 UTC by Gene Czarcinski
Modified: 2005-10-31 22:00 UTC (History)
0 users

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Last Closed: 2003-05-11 15:32:32 UTC

Attachments (Terms of Use)

Description Gene Czarcinski 2003-01-31 16:26:07 UTC
Description of problem:
I have two SMP systems where the install hangs (both gui and text).  I have
tried local cdrom, hard drive, and nfs installs.  Both of these systems have
successfully installed 8.0.92.  The install hangs are various places but usually
when it is starting X.

I suspect that the problem is really a kernel problem but it happens at various
places during install.

The cdroms have been mediachecked OK and successfully used to install on non-smp

Version-Release number of selected component (if applicable):

One system is a dual PIII (933MHz) with 900MB ram and ATI 7500 video.  The other
is a dual PII (400MHz) with 300MB ram and ATI Rage video.

Both successfully install and run  8.0 and 8.0.92.

How reproducible:

Comment 1 Michael Fulbright 2003-01-31 17:34:14 UTC
Anything obviously wrong printed out on any VCs?

Does sysreq tell you anything?

Comment 2 Gene Czarcinski 2003-01-31 18:24:53 UTC
I just ran a test on one of the systems.  I found on VC4 messages saying

  hdd  lost interrupt

multiple times.  hdd is the cdrom drive.  I will check the other system in a
couple of minutes.  Strange that this occurs only with SMP systems.  I also
tried using "linux ide=nodma askmethod" and it hung as soon as it tried to get
to the cdrom.

BTW, I believe the Red Hat mailing list servers are constipated again.

Comment 3 Gene Czarcinski 2003-01-31 18:58:38 UTC
OK, similar thing on the second (dual P-III) system.  Two happenings:

1. Did "linux ide=nodma askmethod" and then selected local cdrom.  This hung
solid  (freeze) when it got the first X screen up with the mouse.  I could do
nothing ... could not switch to VCs.

2. Did "linux ide=nodma askmethod text" and then again selected Local cdrom. 
This got further -- to actual install.  While formating "/" (only one partition
5GB partition defined), it hung about 75% into formatting.  Checking VC4, I found:

Lots of messages saying
  <6>ide:  no cache flush required

and finally:

<3>hdb:  lost interrupt

hdb is the drive containing the "/" partition.

BTW, this system (and the other one) both have 8.0 and 8.0.92 installed and running.

Comment 4 Gene Czarcinski 2003-02-01 18:57:48 UTC
OK, I am changing this to the kernel component.  Thanks to Smooge, the problem
appears to be released to APIC.  I booted the cdrom with "linux noapic" and
successfully installed and ran 8.0.93.

I am ading arjanv to the cc list since that seems to be who get assigned kernel
stuff.  Please adjust the cc list as appropriate or make this a dup of something
that I missed in my searches.

Comment 5 Arjan van de Ven 2003-02-02 10:00:19 UTC
the apic stuff has been removed since a few days; that should fix this too.

Comment 6 Gene Czarcinski 2003-02-02 11:15:53 UTC
I hope just from the install (BOOT) kernel since it seems to work fine when I am
running the smp kernel.

I have not tested it yet but how about the regular non-smp kernel.  Would
running it with APIC enabled screw things up?  If so then maybe it should only
be enabled for the smp kernel.

You comment about APIC being removed a few days ago.  I have not tried one of
the more recent kernels but it is enabled for the 2.4.20-2.24 smp kernel (the
one that does not have the shutdown problem).

Comment 7 Arjan van de Ven 2003-02-02 11:20:27 UTC
yes I meant the BOOT kernel. SMP kernel of course still has apic (no SMP system
without apics exist).

It's removed in -2.27 or so.. -2.30 has the reboot problem fixed fwiw

Note You need to log in before you can comment on or make changes to this bug.