platform/{metal,qemu}: use bootindex=N instead of -boot once=d
#1312
Conversation
If we've injected the config, don't also add `-fw_cfg` even if the platform supports it. This otherwise muddles what we're testing and makes it less obvious that the config must be coming from the injected one.
Makes it easy to double-check that we're compressing the right image (e.g. metal4k and not metal).
When provided, use the metal4k version of the images instead of the regular 512b one.
Let's get some basic coverage of this in cosa. Will add to the FCOS pipeline as well. While we're here, drop the unnecessary `cosa`; kola is cosa workdir aware now.
Symmetrical to `--no-pxe`; skip running the live ISO test if all we want is the PXE test. (The PXE test uses `console=ttyS0` automatically, which makes `--console` more convenient for local testing.)
jlebon
changed the title
-boot once=d
|
Re.
I'm playing with this patch: diff --git a/mantle/platform/metal.go b/mantle/platform/metal.go
index 30f68399..5bd0acd9 100644
--- a/mantle/platform/metal.go
+++ b/mantle/platform/metal.go
@@ -309,7 +309,6 @@ func (inst *Install) setup(kern *kernelSetup) (*installerRun, error) {
pxe.boottype = "pxe"
pxe.networkdevice = "virtio-net-ccw"
pxe.tftpipaddr = "10.0.2.2"
- pxe.bootindex = "1"
default:
return nil, fmt.Errorf("Unsupported arch %s" + system.RpmArch())
}
@@ -440,11 +439,10 @@ func (t *installerRun) completePxeSetup(kargs []string) error {
func (t *installerRun) run() (*QemuInstance, error) {
builder := t.builder
- netdev := fmt.Sprintf("%s,netdev=mynet0,mac=52:54:00:12:34:56", t.pxe.networkdevice)
- if t.pxe.bootindex == "" {
- builder.Append("-boot", "once=n", "-option-rom", "/usr/share/qemu/pxe-rtl8139.rom")
- } else {
- netdev += fmt.Sprintf(",bootindex=%s", t.pxe.bootindex)
+ // re. bootindex=2 arg, see AddInstallIso comment
+ netdev := fmt.Sprintf("%s,netdev=mynet0,mac=52:54:00:12:34:56,bootindex=2", t.pxe.networkdevice)
+ if system.RpmArch() != "s390x" {
+ builder.Append("-option-rom", "/usr/share/qemu/pxe-rtl8139.rom")
}
builder.Append("-device", netdev)
usernetdev := fmt.Sprintf("user,id=mynet0,tftp=%s,bootfile=%s", t.tftpdir, t.pxe.bootfile)Though that doesn't quite work. |
The `-boot once=d` technique only works with some bootloaders. OVMF is not one of those. Instead, do what libvirt also does, which is to use `bootindex=N` flags, which both SeaBIOS and OVMF understand. Though the libvirt VM installation workflow goes one step further here and sets the `bootindex=1` for the CD-ROM only on first boot, then turns off the machine, changes the boot order (and ejects the CD media entirely), then boots it back up. Here instead, we're just relying on the primary disk being unbootable on first boot to avoid having to stop QEMU and starting it again with slightly different args (which is totally possible for us to do, just more complex). This fixes `kola testiso -P --qemu-firmware=uefi` (and by extension, `--qemu-native-4k` too). But note UEFI PXE booting still doesn't work, will try to dig further into that.
Such as whether it was BIOS or UEFI, and whether it was the metal or the metal4k image.
|
Nice! Using the bootindex does make sense to me, particularly if it's what That said, it is more elegant and probably more reliable to eject the CDROM as you mentioned; I think I've seen some installers have a "please eject, then press return" phase which theoretically we could support (and script?). But eh, bootindex works. Hmm...how is the change to use this in CI working without coreos-installer support landed yet? Is it that we're just not doing autodetection and forcing the 4k image to be installed? /lgtm |
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: cgwalters, jlebon The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
Yup, exactly! One can install the 4k images already today. The patch in coreos/coreos-installer#203 only applies to installing from stream metadata. (Which is the default today, but soon won't be once coreos/coreos-installer#197 is merged. But we'll still need the equivalent auto-select logic in osmet too.) |
`kola testiso` knows how to do this now! coreos/coreos-assembler#1312
`kola testiso` knows how to do this now! coreos/coreos-assembler#1312 While we're here, make it run in parallel with the 512b one, and use `kola` directly instead of `cosa kola`.
`kola testiso` knows how to do this now! coreos/coreos-assembler#1312 While we're here, make it run in parallel with the 512b one, and use `kola` directly instead of `cosa kola`.
`kola testiso` knows how to do this now! coreos/coreos-assembler#1312 While we're here, make it run in parallel with the 512b one, and use `kola` directly instead of `cosa kola`.
The
-boot once=dtechnique only works with some bootloaders. OVMF isnot one of those. Instead, do what libvirt also does, which is to use
bootindex=Nflags, which both SeaBIOS and OVMF understand. Though thelibvirt VM installation workflow goes one step further here and sets the
bootindex=1for the CD-ROM only on first boot, then turns off themachine, changes the boot order (and ejects the CD media entirely), then
boots it back up. Here instead, we're just relying on the primary disk
being unbootable on first boot to avoid having to stop QEMU and starting
it again with slightly different args (which is totally possible for us
to do, just more complex).
This fixes
kola testiso -P --qemu-firmware=uefi(and by extension,--qemu-native-4ktoo). But note UEFI PXE booting still doesn't work,will try to dig further into that.