On Thursday, September 24, 2015 03:33:28 PM Håkon Alstadheim wrote: > Den 24. sep. 2015 14:53, skrev Rich Freeman: > > On Thu, Sep 24, 2015 at 6:05 AM, hw <h...@gc-24.de> wrote: > >> Hi, > >> > >> I'm installing Gentoo as a xen PV guest. Do I need to install a > >> bootloader > >> like grub, or should I rather just specify the kernel to boot in the > >> definition file of the guest? If I do the latter, what about the kernel > >> command line? > >> > >> Is there anything I should consider? The host is also running Gentoo. > > > > You can do either. I suspect it would be easier to just use grub. I > > don't know if xen provides a way to provide a command-line, if not you > > could build a default command-line into your kernel. > > Depending on your hardware you will want to use hvm or pvm for > efficiency. (VT-x means hvm is more efficient).
What do you base this on? Without VT-x, HVM doesn't even work, which means PV is only option. With VT-x, PV still has higher performance as the drivers inside the guest talk directly to the host. > If running hvm on > quemu-xen-traditional, you HAVE to use a bootloader inside the VM, or > some kind of netboot/pvgrub thing. If running upstream quemu for a hvm, > you can choose. I find it less of a hassle to use bootloader inside the > VM. It's simple, if you don't have full access to the host. If you have full access, it's actually simpler as you don't have to worry about boot-order, partitioning and a bootloader. > If running pv, I BELIEVE you have to specify kernel and boot options > outside the VM-image, possibly through pvgrub. Pvgrub is supposed to be > able to fish out kernel and initrd from the VM disk. I never bothered to > get that working. Here is an example excerpt from a PVM I use (for a > debian vm): > --print.pvm--- > builder = "generic" > kernel = "/etc/xen/wheezy/vmlinuz-3.16.0-0.bpo.4-amd64" > ramdisk = "/etc/xen/wheezy/initrd.img-3.16.0-0.bpo.4-amd64" > extra = "root=/dev/xvda console=hvc0 tmem" > -- > For explanation check "man xl.cfg". When you specify > kernel&ramdisk¶meters in the vm config, you do NOT need a /boot/ > directory on your vm. Obviously when you build a kernel inside the vm, > it will end up in /boot, so you have to provide a way for the > dom0/hypervisor to load your kernel, either (when using full > virtualization) a regular boot-loader inside the vm, or (for a PV guest) > through some other means to let the dom0/hypervisor load your image. -- Joost
