Bug 200549

Summary: [olpc] i586 kernel panics on boot on AMD Geode
Product: [Fedora] Fedora Reporter: David Zeuthen <davidz>
Component: binutilsAssignee: Jakub Jelinek <jakub>
Status: CLOSED RAWHIDE QA Contact:
Severity: medium Docs Contact:
Priority: high    
Version: rawhideCC: aoliva, davej, hongjiu.lu, mclasen, trever, wtogami
Target Milestone: ---   
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2006-08-01 23:29:35 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On:    
Bug Blocks: 150223, 150224, 191929, 191931    
Attachments:
Description Flags
photograph of panic none

Description David Zeuthen 2006-07-28 16:28:37 UTC 
Description of problem:

panic right after boot - I think even before the initramfs is started. Will
attach screenshot.

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

2.6.17-1.2449.fc6

How reproducible:

Always on AMD Geode. It's really bad as we need to move off the i686 kernel (see
bug 200330 and bug 200409) as the AMD geode is not fully i686 compatible and the
recent binutils started producing code where this is exposed. So right now we're
kind of screwed. Please advise. Thanks!
Comment 1 David Zeuthen 2006-07-28 16:28:39 UTC 
Created attachment 133245 [details]
photograph of panic
Comment 2 Dave Jones 2006-07-28 16:49:27 UTC 
*** Bug 200037 has been marked as a duplicate of this bug. ***
Comment 3 Dave Jones 2006-07-28 17:18:23 UTC 
Ok, this looks like binutils breakage to me. hj 'optimised' the NOP handling for
686, but seems to have busted 586 in the process.

http://sources.redhat.com/ml/binutils/2006-06/msg00193.html

66 0f 1f is not valid on i586 (and possibly some 686s afair).
Comment 4 Dave Jones 2006-07-28 17:31:20 UTC 
ftp://download.intel.com/design/Pentium4/manuals/25366720.pdf states ...

The multi-byte form of NOP is available on processors with model encoding:
ΓΆΒΆ   CPUID.01H.EAX[Bytes 11:8] = 0110B or 1111B

So, only family 6 and 15. Little wonder 586 is hurting.
Comment 5 Dave Jones 2006-07-28 17:42:51 UTC 
note btw, that this statement does nothing to guarantee that the right thing
happens with that code sequence on for eg, athlons, VIA-C3, or any other
non-intel 686.  It seems Intel only recently started documenting the multi-byte
variants of NOP, so any vendors that based their own 686 implementation on
reading (older) intel docs may be in for a surprise.
Comment 6 Christopher Blizzard 2006-07-28 19:38:59 UTC 
So what's the plan to fix things here?  Unscrew binutils?  In our example it
would be great if we can compile somewhere else (for i686) and it still runs on
the geode.  Is upstream binutils actually broken in this case?
Comment 7 H.J. Lu 2006-07-28 22:50:52 UTC 
Can you try this patch

http://sourceware.org/ml/binutils/2006-07/msg00360.html
Comment 8 Christopher Blizzard 2006-08-01 22:58:05 UTC 
Sounds like we have a working kernel after a rebuild with binutils and this
patch.  David, are we good to close?
Comment 9 David Zeuthen 2006-08-01 23:29:35 UTC 
Yup, the i686 kernel works on the Geode. Closing bug. Thanks to everyone involved!