Re: d-i support for running in a Xen guest domain
On Sat, 2008-01-12 at 17:30 +, Ian Campbell wrote: Unfortunately the Xen domain builder isn't capable of loading a native bzImage directly -- it requires the ELF vmlinux. I hacked around that when I was playing with Xen enabled d-i and then promptly forgot I had done so, which is a shame because it's quite important! For now I grabbed the 686-bigmem vmlinux from the build tree and boot tested that. Took me a while but there is now a patch queued with upstream which fixes this issue http://git.kernel.org/?p=linux/kernel/git/x86/linux-2.6-x86.git;a=commit;h=23eea6d21020dfaa740c01b911f46a95505c9b37 would a backport to 2.6.24 be applied by Debian? I'm working on the domain builder bits on the Xen side at the moment. Ian. -- Ian Campbell When in doubt, mumble; when in trouble, delegate; when in charge, ponder. -- James H. Boren signature.asc Description: This is a digitally signed message part
Re: d-i support for running in a Xen guest domain
On Sat, 12 Jan 2008, Ian Campbell wrote: Unfortunately the Xen domain builder isn't capable of loading a native bzImage directly -- it requires the ELF vmlinux. I hacked around that when I was playing with Xen enabled d-i and then promptly forgot I had done so, which is a shame because it's quite important! For now I grabbed the 686-bigmem vmlinux from the build tree and boot tested that. hmm there were the xen images i guess they did something similar, maybe look there on how to grab them with maintainer scripts. so enabling won't bring us much for now also due to !pae i only see a current value in i386 bigmem. The Fedora guys announced that they were working on dom0 paravirt_ops stuff at the end of last year [0]. I must admit I haven't really been keeping up with their efforts though. I had a dig around and didn't see anything further to what was announced. They were targeting Fedora 9. According to [1] feature freeze is start of March so I'd presume they plan to have patches before then. referring to that: http://fedoraproject.org/wiki/Features/XenPvops (seem to not load right now on my end, but should be the page) amicalement -- maks -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: d-i support for running in a Xen guest domain
On Sat, Jan 12, 2008 at 05:30:20PM +, Ian Campbell wrote: The explicit Xen images will disappear in the future, it just lacks infrastructure. By infrastructure do you mean meta-packages to facilitate upgrade or something else? Infrastructure as in maintainer scripts. Bastian -- Killing is wrong. -- Losira, That Which Survives, stardate unknown -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: d-i support for running in a Xen guest domain
On Thu, 2008-01-10 at 13:37 +0100, Bastian Blank wrote: On Wed, Jan 09, 2008 at 09:58:45PM +, Ian Campbell wrote: I'm interested in getting d-i to work seamlessly inside a Xen guest domain using the paravirt_ops kernel support for Xen in 2.6.24+ (currently only available in the kernel svn trunk snapshots builds I think). 2.6.23 already have the same amount of support than .24. That's true. I'm not sure why I thought otherwise. It would be reasonably easy to create -xen-686 udebs (in fact I did for my testing) and tweak d-i to select a -xen kernel package image as appropriate during install, however I think I have a better idea ;-)... The explicit Xen images will disappear in the future, it just lacks infrastructure. By infrastructure do you mean meta-packages to facilitate upgrade or something else? Ian. -- Ian Campbell And ever has it been known that love knows not its own depth until the hour of separation. -- Kahlil Gibran signature.asc Description: This is a digitally signed message part
Re: d-i support for running in a Xen guest domain
On Thu, 2008-01-10 at 00:59 +0100, maximilian attems wrote: On Wed, Jan 09, 2008 at 09:58:45PM +, Ian Campbell wrote: snipp d-i side I'd like to propose enabling Xen guest support in all the native i?86 kernel images (and eventually amd64 too), or at least in the -686-bigmem kernel. Since paravirt ops is already enabled in all kernels (for KVM and lguest) there is only minimal additional overhead to enabling Xen. As a nice bonus this would also allow the separate -xen flavour kernels to be removed/deprecated which seems like a good thing in the paravirt_ops based world. this sounds like a good plan, getting rid of the number of images is always a welcome bonus. I have build tested the attached patch and test booted the -486 and -686-bigmem versions on native. Unfortunately the Xen domain builder isn't capable of loading a native bzImage directly -- it requires the ELF vmlinux. I hacked around that when I was playing with Xen enabled d-i and then promptly forgot I had done so, which is a shame because it's quite important! For now I grabbed the 686-bigmem vmlinux from the build tree and boot tested that. People were looking at modifications to the kernel and/or the Xen domain builder to allow it to boot a bzImage, I think by making the bzImage payload be an ELF file instead of the current raw binary blob. I'll see what is going on in that area. I don't think there is any harm in applying the patch now, although without a patch for builder support it's a bit pointless. i know that fedora also is very keen to move to the paravirt_ops world. for Etch linux-2.6 had dom0 images, when will those be merged, where are the patches based on paravirt ops against 2.6.24? we wouldn't want to release Lenny without the support we had given in Etch (which turned out to be *very* buggy, but that is another story) of a dom0 linux-2.6 linux-image variant. The Fedora guys announced that they were working on dom0 paravirt_ops stuff at the end of last year [0]. I must admit I haven't really been keeping up with their efforts though. I had a dig around and didn't see anything further to what was announced. They were targeting Fedora 9. According to [1] feature freeze is start of March so I'd presume they plan to have patches before then. [0] https://www.redhat.com/archives/fedora-xen/2007-November/msg00106.html [1] http://fedoraproject.org/wiki/Releases/9 Cheers, Ian. -- Ian Campbell From the cradle to the coffin underwear comes first. -- Bertolt Brecht Index: linux-2.6/debian/config/i386/config === --- linux-2.6.orig/debian/config/i386/config 2008-01-10 19:05:40.0 + +++ linux-2.6/debian/config/i386/config 2008-01-10 19:06:19.0 + @@ -1483,7 +1483,10 @@ CONFIG_KS0108_DELAY=2 CONFIG_CFAG12864B=m CONFIG_CFAG12864B_RATE=20 -# CONFIG_XEN is not set +CONFIG_XEN=y +CONFIG_XEN_BLKDEV_FRONTEND=y +CONFIG_HVC_XEN=y +CONFIG_XEN_NETDEV_FRONTEND=y CONFIG_VMI=y CONFIG_PARAVIRT=y CONFIG_PARAVIRT_GUEST=y Index: linux-2.6/debian/changelog === --- linux-2.6.orig/debian/changelog 2008-01-10 19:05:40.0 + +++ linux-2.6/debian/changelog 2008-01-10 19:06:19.0 + @@ -63,6 +63,9 @@ [ Christian T. Steigies ] * [m68k]: Update patches from linux-m68k CVS + [ Ian Campbell ] + * [i386]: Enable Xen guest support in all i386 flavours. + -- maximilian attems [EMAIL PROTECTED] Wed, 09 Jan 2008 17:17:56 +0100 linux-2.6 (2.6.23-1~experimental.1) UNRELEASED; urgency=low signature.asc Description: This is a digitally signed message part
Re: d-i support for running in a Xen guest domain
On Wed, Jan 09, 2008 at 09:58:45PM +, Ian Campbell wrote: I'm interested in getting d-i to work seamlessly inside a Xen guest domain using the paravirt_ops kernel support for Xen in 2.6.24+ (currently only available in the kernel svn trunk snapshots builds I think). 2.6.23 already have the same amount of support than .24. It would be reasonably easy to create -xen-686 udebs (in fact I did for my testing) and tweak d-i to select a -xen kernel package image as appropriate during install, however I think I have a better idea ;-)... The explicit Xen images will disappear in the future, it just lacks infrastructure. Bastian -- Yes, it is written. Good shall always destroy evil. -- Sirah the Yang, The Omega Glory, stardate unknown -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: d-i support for running in a Xen guest domain
On Wed, Jan 09, 2008 at 09:58:45PM +, Ian Campbell wrote: snipp d-i side I'd like to propose enabling Xen guest support in all the native i?86 kernel images (and eventually amd64 too), or at least in the -686-bigmem kernel. Since paravirt ops is already enabled in all kernels (for KVM and lguest) there is only minimal additional overhead to enabling Xen. As a nice bonus this would also allow the separate -xen flavour kernels to be removed/deprecated which seems like a good thing in the paravirt_ops based world. this sounds like a good plan, getting rid of the number of images is always a welcome bonus. i know that fedora also is very keen to move to the paravirt_ops world. for Etch linux-2.6 had dom0 images, when will those be merged, where are the patches based on paravirt ops against 2.6.24? we wouldn't want to release Lenny without the support we had given in Etch (which turned out to be *very* buggy, but that is another story) of a dom0 linux-2.6 linux-image variant. Given a suitable kernel it would then be possible to add a -bigmem flavour of d-i. That would give us installer images which support native, Xen, KVM and lguest via paravirt_ops both PAE and non-PAE allowing Debian to be installed as a guest on a wide variety of domain0 distributions. linux-2.6 already builds the -bigmem flavour and afaik it is instelled by d-i later but not used on boot. So, what do the kernel folks think about enabling Xen guest support in all kernels where it is available and getting rid of the -xen specific variants? I've got a massive (mainly due to lots of deletion) patch against SVN which does just that, if there is interest I'll work out the kinks and file it as a wishlist bug. 2.6.24 is right now open for such changes, thanks for your already filed and fixed virtualization i386 bug report, had gone unnoticed for a while as focus slipps to amd64. happy hacking -- maks -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
