Tom Brown
2006-Aug-17 08:00 UTC
Re: [Xen-users] hvm, handle_mmio: failed to get instruction length
Is there a problem booting 2.4 kernels compiled on Redhat 7 machines? After some sweat and cursing I''ve gotten a config where grub works reliably, and I can get a 2.6 kernel to boot before it spills its guts all over, but I can not get an old 2.4 kernel to boot ... I should dig out a 2.2 kernel, as that actually would have a purpose, but I was under the belief (possibly/probably self delusion) that running linux kernels with HVM would be easy... I''ve tried xen-3.0.2-2 and -unstable and now -testing... all with the same result. With unstable I did find a patch for can not find an E820 memory map... but I have not been able to find a fix for this one... (XEN) Code bytes around(len=256) c01fc909: (XEN) c01fc900:0f 0d 86 40 01 00 00 0f 6f 06 0f 6f 4e 08 0f 6f (XEN) c01fc910:56 10 0f 6f 5e 18 0f 7f 03 0f 7f 4b 08 0f 7f 53 (XEN) c01fc920:10 0f 7f 5b 18 0f 6f 46 20 0f 6f 4e 28 0f 6f 56 (XEN) c01fc930:30 0f 6f 5e 38 0f 7f 43 20 0f 7f 4b 28 0f 7f 53 (XEN) c01fc940:30 0f 7f 5b 38 4f 83 c6 40 83 c3 40 83 ff 05 7f (XEN) c01fc950:af eb 45 0f 6f 06 0f 6f 4e 08 0f 6f 56 10 0f 6f (XEN) c01fc960:5e 18 0f 7f 03 0f 7f 4b 08 0f 7f 53 10 0f 7f 5b (XEN) c01fc970:18 0f 6f 46 20 0f 6f 4e 28 0f 6f 56 30 0f 6f 5e (XEN) c01fc980:38 0f 7f 43 20 0f 7f 4b 28 0f 7f 53 30 0f 7f 5b (XEN) c01fc990:38 4f 83 c6 40 83 c3 40 85 ff 7f b7 89 e8 83 e0 (XEN) c01fc9a0:3f 89 c1 c1 e9 02 89 df f3 a5 a8 02 74 02 66 a5 (XEN) c01fc9b0:a8 01 74 01 a4 0f 20 c0 83 c8 08 0f 22 c0 8b 04 (XEN) c01fc9c0:24 5a 5b 5e 5f 5d c3 89 f6 8d bc 27 00 00 00 00 (XEN) c01fc9d0:53 8b 5c 24 08 e8 d6 f6 f0 ff 0f ef c0 b8 3f 00 (XEN) c01fc9e0:00 00 8d b4 26 00 00 00 00 8d bc 27 00 00 00 00 (XEN) c01fc9f0:0f e7 03 0f e7 43 08 0f e7 43 10 0f e7 43 18 0f (XEN) handle_mmio: failed to get instruction length (XEN) domain_crash_sync called from platform.c:846 (XEN) Domain 12 (vcpu#0) crashed on cpu#1: any thoughts? AFAIK, there is nothing magic here. I can post configs if needed, but the dump seems like it is supposed to be sufficient. No, I have not compared all the hex dumps to see if they are consistent... hhmm, that probably is a good idea... yes, it looks quite repeatable. -Tom _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Petersson, Mats
2006-Aug-17 09:22 UTC
RE: [Xen-users] hvm, handle_mmio: failed to get instruction length
> -----Original Message----- > From: xen-users-bounces@lists.xensource.com > [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of Tom Brown > Sent: 17 August 2006 09:01 > To: xen-users@lists.xensource.com > Subject: Re: [Xen-users] hvm, handle_mmio: failed to get > instruction length > > > Is there a problem booting 2.4 kernels compiled on Redhat 7 machines? > After some sweat and cursing I''ve gotten a config where grub works > reliably, and I can get a 2.6 kernel to boot before it spills > its guts all > over, but I can not get an old 2.4 kernel to boot ... I > should dig out a > 2.2 kernel, as that actually would have a purpose, but I was > under the > belief (possibly/probably self delusion) that running linux > kernels with > HVM would be easy... > > I''ve tried xen-3.0.2-2 and -unstable and now -testing... all > with the same > result. With unstable I did find a patch for can not find an > E820 memory > map... but I have not been able to find a fix for this one... > > (XEN) Code bytes around(len=256) c01fc909: > (XEN) c01fc900:0f 0d 86 40 01 00 00 0f 6f 06 0f 6f 4e 08 0f 6fI bet your kernel is compiled for "K7" or some such, right? It looks like you''re trying to perform a 3DNow!/MMX/SSE instruction, and that is known to not work... Try building/using a 2.4 kernel compiled for "plain" i686 with no fancy features. [It''s faster anyways, since it will use the REP; MOVS instruction to copy memory, which will intercept ONCE, and do all the work inside Xen, then return, rather than multiple intercepts - as opposed on real hardware were each read/write operation is the critical path and getting more data into one write is helpful]. Appologies if my decoding of instructions is wrong, and you''re actually running a plain i686/i386 kernel - in that case, please reply again, and I''ll dig out the opcode list and try to figure out what goes on. By the way, it''s entirely my fault that this doesn''t work - I was supposed to fix it, and I took a week''s holiday when I should have been working on it - hence I couldn''t get a patch ready for the end of June - July 1st I was put on a different project. My sincere appologies... -- Mats> (XEN) c01fc910:56 10 0f 6f 5e 18 0f 7f 03 0f 7f 4b 08 0f 7f 53 > (XEN) c01fc920:10 0f 7f 5b 18 0f 6f 46 20 0f 6f 4e 28 0f 6f 56 > (XEN) c01fc930:30 0f 6f 5e 38 0f 7f 43 20 0f 7f 4b 28 0f 7f 53 > (XEN) c01fc940:30 0f 7f 5b 38 4f 83 c6 40 83 c3 40 83 ff 05 7f > (XEN) c01fc950:af eb 45 0f 6f 06 0f 6f 4e 08 0f 6f 56 10 0f 6f > (XEN) c01fc960:5e 18 0f 7f 03 0f 7f 4b 08 0f 7f 53 10 0f 7f 5b > (XEN) c01fc970:18 0f 6f 46 20 0f 6f 4e 28 0f 6f 56 30 0f 6f 5e > (XEN) c01fc980:38 0f 7f 43 20 0f 7f 4b 28 0f 7f 53 30 0f 7f 5b > (XEN) c01fc990:38 4f 83 c6 40 83 c3 40 85 ff 7f b7 89 e8 83 e0 > (XEN) c01fc9a0:3f 89 c1 c1 e9 02 89 df f3 a5 a8 02 74 02 66 a5 > (XEN) c01fc9b0:a8 01 74 01 a4 0f 20 c0 83 c8 08 0f 22 c0 8b 04 > (XEN) c01fc9c0:24 5a 5b 5e 5f 5d c3 89 f6 8d bc 27 00 00 00 00 > (XEN) c01fc9d0:53 8b 5c 24 08 e8 d6 f6 f0 ff 0f ef c0 b8 3f 00 > (XEN) c01fc9e0:00 00 8d b4 26 00 00 00 00 8d bc 27 00 00 00 00 > (XEN) c01fc9f0:0f e7 03 0f e7 43 08 0f e7 43 10 0f e7 43 18 0f > (XEN) handle_mmio: failed to get instruction length > (XEN) domain_crash_sync called from platform.c:846 > (XEN) Domain 12 (vcpu#0) crashed on cpu#1: > > > any thoughts? AFAIK, there is nothing magic here. I can post > configs if > needed, but the dump seems like it is supposed to be > sufficient. No, I > have not compared all the hex dumps to see if they are > consistent... hhmm, > that probably is a good idea... yes, it looks quite repeatable. > > -Tom > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Tom Brown
2006-Aug-17 16:57 UTC
RE: [Xen-users] hvm, handle_mmio: failed to get instruction length
> > > > I''ve tried xen-3.0.2-2 and -unstable and now -testing... all > > with the same > > result. With unstable I did find a patch for can not find an > > E820 memory > > map... but I have not been able to find a fix for this one... > > > > (XEN) Code bytes around(len=256) c01fc909: > > (XEN) c01fc900:0f 0d 86 40 01 00 00 0f 6f 06 0f 6f 4e 08 0f 6f > > I bet your kernel is compiled for "K7" or some such, right? It looks > like you''re trying to perform a 3DNow!/MMX/SSE instruction, and that is > known to not work...THANK YOU! That was it. Most of our older servers were Athlon MP or XP based, so we''ve got a lot of K7 configured kernels. Hopefully this thread will be of help to more folks than just me. As for it being your fault, that isn''t true... it would be the fault of whomever re-assigned you. -Tom _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users