I''m trying to write the definitive 32 and 64 answer for the FAQ, but I''m missing quite a bit. I''ve posted what I have: http://wiki.xensource.com/xenwiki/XenFaq#head-5f7176b3909cb0382cece43a6a8fc25a3a114e93 http://tinyurl.com/2otmeb First, if I''m just plain wrong about anything _please_ let me know or just fix the FAQ entry. Here''s some questions: 1) What really is 32-on-64? From what I''ve read I should be able to run 32bit PV domUs on a 64bit dom0 with 3.1.0. I''ve tried it and it only works with a 64 bit kernel. What am I missing? 2) Can you run 64 bit domUs on a 32bit guest in any way? 3) Can I just assume that people are using the same bitness in hardware, hypervisor and dom0 kernel and userland? Can I consider someone running a 32bit hypervisor and dom0 kernel and userland to just be the same as someone using 32bit hardware? Thanks! -Dylan _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Saturday 04 August 2007, Dylan Martin wrote:> 1) What really is 32-on-64? From what I''ve read I should be able to > run 32bit PV domUs on a 64bit dom0 with 3.1.0. I''ve tried it and it > only works with a 64 bit kernel. What am I missing?As it says on the wiki the 32bit PV kernel is required to have PAE. Did you remember that? What error did you get? I just tried it out with dom0 64bit and domU 32bit PAE On an AMD single core system(64 3700+) it worked just fine and it runs 32 bit PAE and 64bit domUs at the same time. On an Intel dual dual core system (2x5120) however the 32bit PAE kernel panics before mounting the root filesystem. Something about not being able to read superblock on xvda1. All of the kernels are compiled from xensource 3.1 except the intel system which uses 2.6.20 from gentoo xen overlay. So maybe that is the problem. And what I would like to know is which combination will perform best of the 2 choices of 32bit userland/64bit kernel and 32bit userland/32bit kernel both on a 64 bit dom0. And are there other advantages/disadvantages of the 2 choices. -- Jesper 20:20:52 up 1 day, 1:25, 11 users, load average: 0.12, 0.07, 0.01 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> On Saturday 04 August 2007, Dylan Martin wrote: > > 1) What really is 32-on-64? From what I''ve read I should be able to > > run 32bit PV domUs on a 64bit dom0 with 3.1.0. I''ve tried it and it > > only works with a 64 bit kernel. What am I missing? > > As it says on the wiki the 32bit PV kernel is required to have PAE. Did > you remember that? What error did you get?I might have forgotten to use a PAE kernel, but see below.> I just tried it out with dom0 64bit and domU 32bit PAE > On an AMD single core system(64 3700+) it worked just fine and it runs 32 > bit PAE and 64bit domUs at the same time. > On an Intel dual dual core system (2x5120) however the 32bit PAE kernel > panics before mounting the root filesystem. Something about not being > able to read superblock on xvda1.Ah ha! My hardware is Intel dual-core and I had panics from xvda1 too. Sounds like a bug!> All of the kernels are compiled from xensource 3.1 except the intel system > which uses 2.6.20 from gentoo xen overlay. So maybe that is the problem.Mine was Fedora 7 and I had the same error.> And what I would like to know is which combination will perform best of > the 2 choices of 32bit userland/64bit kernel and 32bit userland/32bit > kernel both on a 64 bit dom0. And are there other > advantages/disadvantages of the 2 choices. > > -- > Jesper > 20:20:52 up 1 day, 1:25, 11 users, load average: 0.12, 0.07, 0.01 > > _______________________________________________ > 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
On Tuesday 07 August 2007, Dylan Martin wrote:> Ah ha! My hardware is Intel dual-core and I had panics from xvda1 too. > Sounds like a bug! > > > All of the kernels are compiled from xensource 3.1 except the intel > > system which uses 2.6.20 from gentoo xen overlay. So maybe that is > > the problem. > > Mine was Fedora 7 and I had the same error.The gentoo 2.6.20 kernel is based on Fedora 7. So it might just be the Fedora kernel that has this problem. I will try the xensource kernel later when the computer is not busy. -- Jesper 09:09:06 up 4:14, 4 users, load average: 0.72, 0.74, 1.08 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
The error I get is like this. I guess the error is actually not reading xvda1, but the next line..? Is it the same you get? ----- [325504.445871] XENBUS: Device with no driver: device/console/0 [325504.446435] end_request: I/O error, dev xvda1, sector 16 [325504.446450] ReiserFS: xvda1: warning: sh-2006: read_super_block: bread failed (dev xvda1, block 2, size 4096) [325504.446483] BUG: unable to handle kernel paging request at virtual address d9c2dce4 [325504.446490] printing eip: [325504.446493] c023236a [325504.446497] 003f6000 -> *pde = 00000000:06152027 [325504.446501] 003f7000 -> *pme = 00000000:00000000 [325504.446506] Oops: 0000 [#1] [325504.446511] CPU: 0 [325504.446512] EIP: 0061:[<c023236a>] Not tainted VLI [325504.446513] EFLAGS: 00010887 (2.6.18 #3) [325504.446525] EIP is at blkif_int+0x8f/0x1e1 [325504.446529] eax: 189c9c00 ebx: c1264000 ecx: 00000000 edx: 00000002 [325504.446534] esi: c125e0ac edi: 00000000 ebp: ca010100 esp: c0389f1c [325504.446538] ds: 007b es: 007b ss: e021 [325504.446543] Process swapper (pid: 0, ti=c0388000 task=c03322a0 task.ti=c0388000) [325504.446547] Stack: 0021caa0 00000103 00000001 00000002 00000000 00000000 00000001 c121cac0 [325504.446560] 00000000 00000000 00000103 c0128ee5 c0389fa0 c03830c0 00000103 c121cac0 [325504.446573] c0389fa0 c0128f6e 00000103 c0389fa0 f57ef000 fffffffe c0105e28 00000000 [325504.446586] Call Trace: [325504.446590] [<c0128ee5>] handle_IRQ_event+0x35/0x63 [325504.446597] [<c0128f6e>] __do_IRQ+0x5b/0x9f [325504.446602] [<c0105e28>] do_IRQ+0x66/0x76 [325504.446608] [<c0229dc2>] evtchn_do_upcall+0x5d/0x98 [325504.446614] [<c0104955>] hypervisor_callback+0x3d/0x45 [325504.446620] [<c0106dfb>] raw_safe_halt+0xa0/0xbc [325504.446626] [<c0103873>] xen_idle+0x34/0x40 [325504.446631] [<c0102cfa>] cpu_idle+0x2c/0x41 [325504.446636] [<c038a628>] start_kernel+0x2a4/0x2a6 [325504.446642] Code: 8b 43 20 89 44 24 08 89 44 24 18 e9 f3 00 00 00 8b 43 24 31 ff 48 23 44 24 08 6b c0 6c 8d 70 40 03 73 28 8b 2e 69 c5 9c 00 00 00 <8b> 94 18 e4 00 00 00 8d 44 18 78 89 44 24 14 89 54 24 04 eb 12 [325504.446703] EIP: [<c023236a>] blkif_int+0x8f/0x1e1 SS:ESP e021:c0389f1c [325504.446711] <0>Kernel panic - not syncing: Fatal exception in interrupt -- Jesper 09:38:48 up 4:44, 5 users, load average: 0.47, 0.59, 0.65 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Wed, Aug 08, 2007 at 10:09:30AM +0800, Jesper Andersen wrote:> The error I get is like this. > I guess the error is actually not reading xvda1, but the next line..? > Is it the same you get? > > ----- > > [325504.445871] XENBUS: Device with no driver: device/console/0 > [325504.446435] end_request: I/O error, dev xvda1, sector 16 > [325504.446450] ReiserFS: xvda1: warning: sh-2006: read_super_block: bread > failed (dev xvda1, block 2, size 4096) > [325504.446483] BUG: unable to handle kernel paging request at virtual > address d9c2dce4 > [325504.446490] printing eip: > [325504.446493] c023236a > [325504.446497] 003f6000 -> *pde = 00000000:06152027 > [325504.446501] 003f7000 -> *pme = 00000000:00000000 > [325504.446506] Oops: 0000 [#1] > [325504.446511] CPU: 0 > [325504.446512] EIP: 0061:[<c023236a>] Not tainted VLI > [325504.446513] EFLAGS: 00010887 (2.6.18 #3) > [325504.446525] EIP is at blkif_int+0x8f/0x1e1 >[...]Getting the same on fedora7 here too trying to boot the rhel4.5 and rhel5 paravirtualized installkernels. And im pretty sure the xen-config is correct since the 32bit-installkernels of both distros boot and install the domU just fine. Christian _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
I still didn''t try another kernel on that server. But I tried the xensource 3.1 on an Intel 6300 cpu system running kubuntu and here it worked fine with 32bit pae domU. So I guess it is the Fedora 7 kernel that has a problem. -- Jesper 23:57:56 up 2:20, 4 users, load average: 0.59, 0.81, 0.79 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
And I finally changed the kernel on the server with the 2 5120 to the xensource 3.1. Now 32bit PAE domU is working on that as well. -- Jesper 08:28:12 up 3:51, 6 users, load average: 0.33, 0.37, 0.53 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Just one question, how do you specify in your xen.cfg file to pick up the 32-bit kernel and start the xen instance? Right now I am doing the following: kernel = "/boot/vmlinuz-2.6-xen" ramdisk = "/boot/initrd-2.6.18-xen.img" But this seems to be picking up the 64-bit xensource kernel in my dom0''s /boot directory, not the identically-named /boot/vmlinux-2.6-xen in what is going to be the /boot partition of dom1 (which I have populated with 32-bit rhel5). Do I have to have the 32-bit kernel that the domU is going to boot, visible to user space within the dom0 so xen can start the instance properly? What is happening now is that the domU is starting and I am ending up running a 64-bit kernel on top of a 32-bit distro, which I thought was impossible but it is mostly working. Any help is appreciated. Thanks Steve Timm On Fri, 10 Aug 2007, Jesper Andersen wrote:> I still didn''t try another kernel on that server. But I tried the > xensource 3.1 on an Intel 6300 cpu system running kubuntu and here it > worked fine with 32bit pae domU. So I guess it is the Fedora 7 kernel > that has a problem. > >-- ------------------------------------------------------------------ Steven C. Timm, Ph.D (630) 840-8525 timm@fnal.gov http://home.fnal.gov/~timm/ Fermilab Computing Division, Scientific Computing Facilities, Grid Facilities Department, FermiGrid Services Group, Assistant Group Leader. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Wednesday 15 August 2007, Steven Timm wrote:> Just one question, how do you specify in your xen.cfg > file to pick up the 32-bit kernel and start the xen instance? > Right now I am doing the following: > > kernel = "/boot/vmlinuz-2.6-xen" > > ramdisk = "/boot/initrd-2.6.18-xen.img" > > But this seems to be picking up the 64-bit xensource kernel > in my dom0''s /boot directory, not the identically-named > /boot/vmlinux-2.6-xen in what is going to be the /boot partition > of dom1 (which I have populated with 32-bit rhel5).When specified this way kernels are picket from dom0. So to make it work you will have to copy the kernel to dom0. If you want to keep the kernel in domU you can use pygrub to write something like grubs menu.lst to choose kernel.> Do I have to have the 32-bit kernel that the domU is going to > boot, visible to user space within the dom0 so xen can start the > instance properly?When you specify this way you have to.> What is happening now is that the domU is starting and I am > ending up running a 64-bit kernel on top of a 32-bit distro, > which I thought was impossible but it is mostly working.64 bit kernels can have support for executing 32 bit code. I have been using that for a long time. -- Jesper 11:04:10 up 13:58, 5 users, load average: 0.60, 0.28, 0.11 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users