Mail archive
alpine-devel

Re: [alpine-devel] Alpine Linux 3.7.0 Release Candidate 1 on VirtualBox

From: Kiyoshi Aman <aphrael_at_alpinelinux.org>
Date: Fri, 01 Dec 2017 16:49:24 +0000

I have hardware which will not boot in 'legacy BIOS' mode at all (Atom
x5-Z8330-based Intel compute stick), and while VMs don't necessarily need
to emulate that, it can be helpful to force EFI-only boot for debugging
purposes.

On Fri, Dec 1, 2017 at 10:44 AM Jakub Jirutka <jakub_at_jirutka.cz> wrote:

> Hi Gennady,
>
> Any reason you are against enabling it?
>
>
> I’m not strongly against it, I just prefer to keep number of enabled
> features at minimum (and yes, I know that the current configs are exactly
> opposite). I don’t see any added value of EFI for VMs. Actually I can
> hardly see it even for physical HW, it just allowed vendors to add more
> crap and bloat into it, but that’s a different story.
>
> It other devs don’t mind adding these features, let’s do it.
>
> Jakub
>
> On 1. Dec 2017, at 16:10, Gennady Feldman <gena01_at_gmail.com> wrote:
>
> Hi Jakub,
>
> Was thinking a bit more about virt + EFI.. Any reason you are against
> enabling it? It looks like various VM platforms have already added or are
> adding EFI support to replace "legacy BIOS" as part of the VM environment.
>
> Right now that kernel doesn't boot and from the looks of it it's only
> enabling 2 modules.
>
> P.S. Can somebody bump community/virtualbox-guest-additions/APKBUILD
> 5.1.26 => 5.1.30?
>
> Thank you,
>
> Gennady
>
> On Thu, Nov 30, 2017 at 7:50 PM, Gennady Feldman <gena01_at_gmail.com> wrote:
>
>> HI Jakub,
>>
>> I am going to answer what I know ( I am sure others will have their own
>> reasons). I have Virtualbox 5.1.30 (there's also 5.2.x out) installed on
>> Windows 7.
>> Right now from what I can see it's NOT a requirement, but the option is
>> there.
>>
>> I did download and try to play a bit with the 3.7rc3 both virt and
>> standard. Virt doesn't even boot. Standard seems to be fine as long as I
>> don't install the virt kernel and try to boot with it.
>>
>> From what I read EFI seems like the replacement for BIOS and should
>> provide a number of benefits including a better/bigger console (which I
>> like quite a bit, kinda hate doing console installs
>> using legacy BIOS inside VirtualBox) and faster boot time among other
>> things. I might even convert my Alpine VMs to use it once virt kernel
>> supports it.
>>
>> I guess from MBR <=> GPT perspective it could go either way. Things might
>> get interesting with bigger external volumes, but not for something I am
>> working/messing with right now.
>>
>> setup-disk does some interesting stuff in EFI mode:
>> - boot has hard-coded 512 (which seems quite high?)
>> - creates a swap partition (2G for me). I would love to turn that off if
>> I am running inside VirtualBox or being able to choose.
>>
>> I also wanted to ask if it would make sense to have "setup-virtualbox"
>> script that would install virtualbox packages and tweak some things for VMs
>> running inside VirtualBox.
>>
>> Thank you,
>>
>> Gennady
>>
>> On Thu, Nov 30, 2017 at 9:52 AM, Jakub Jirutka <jakub_at_jirutka.cz> wrote:
>>
>>> Wait, VirtualBox *requires* EFI now? If not, why to enable it in kernel
>>> for VMs?
>>>
>>> Jakub
>>>
>>> On 30. Nov 2017, at 15:50, Gennady Feldman <gena01_at_gmail.com> wrote:
>>>
>>> Hi,
>>>
>>> From a quick look looks like:
>>> CONFIG_EFI=y
>>> CONFIG_EFI_STUB=y
>>>
>>> are missing from the virthardened config.
>>>
>>> Thank you,
>>>
>>> On Wed, Nov 29, 2017 at 12:14 PM, Natanael Copa <ncopa_at_alpinelinux.org>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> Thanks you very much for testing.
>>>>
>>>> I think the most critical issues are fixed. I have not bothered fix the
>>>> eject issue, since I think it mostly affects virtualbox only.
>>>>
>>>> I haven't looked at the alpine-virt + efi issue, yet. I suppose there
>>>> are some kernel modules that are needed?
>>>>
>>>> Can you please re-run the tests with the 3.7.0_rc3?
>>>>
>>>> Thanks!
>>>>
>>>> -nc
>>>>
>>>> On Sat, 25 Nov 2017 21:39:02 -0500
>>>> Jack Schmidt <alpine_at_mowsey.org> wrote:
>>>>
>>>> > Here are my notes on Alpine 3.7.0 rc1 installations using VirtualBox
>>>> hosted on x86_64 Darwin.
>>>> >
>>>> >
>>>> > Summary:
>>>> >
>>>> > * Standard, Extended, Vanilla, Virtual on BIOS just work (minor
>>>> issues)
>>>> > * Standard, Extended, Vanilla on EFI has one major issue (bad fsck)
>>>> > * Virtual is broken on EFI (no kernel support)
>>>> >
>>>> > In each case I tried to do a base install to an IDE disk, as "sys",
>>>> default options. I also tried booting the ISO as a run-from-ram system.
>>>> Virtual on EFI requires a serial console, but otherwise all ISOs
>>>> run-from-ram smoothly.
>>>> >
>>>> > The fsck issue needs to be fixed. The 4.9.63 vs 4.9.59 should be
>>>> fixed. The Virtual EFI probably should be fixed, but is easily
>>>> worked-around by end users. The minor issues would be nice to increase
>>>> adoption of bootable alpine, but aren't necessary. I assume docker
>>>> containers are just fine: I use edge all the time without trouble.
>>>> >
>>>> >
>>>> > Standard, Extended, Vanilla, Virtual legacy BIOS minor issues:
>>>> >
>>>> > * Install boot might be too quiet for too long
>>>> >
>>>> > The BIOS boot takes 5 to 20 seconds, so the "quiet" option in
>>>> boot/syslinux/syslinux.cfg might need to be accompanied by a "say" or a
>>>> "menu autoboot" type thing. Grub EFI boot is mostly better, but sometimes
>>>> takes a long time between messages. The installed version is a little slow,
>>>> but has a reasonable amount of messages to let you know it is progressing.
>>>> >
>>>> >
>>>> > Standard, Extended, Vanilla, Virtual both BIOS and EFI minor issues:
>>>> >
>>>> > * After setup-alpine is complete, eject cdrom
>>>> > * Install boot and installed root use different kernels, rescue is
>>>> hard
>>>> > * (hardened/virthardened) Guest additions kernel version problem
>>>> >
>>>> > Install for BIOS works easily on Standard, Extended, Vanilla,
>>>> Virtual. Install for EFI works easily on Standard, Extended, Vanilla, but
>>>> fails for Virtual.
>>>> >
>>>> > After the install, setup-alpine asks to reboot, but doesn't "eject
>>>> -s" (-s required for me) the cd. That means alpine reboots under the
>>>> readonly cdrom with a tmpfs. This works fine, but is surprising if you are
>>>> installing software.
>>>> >
>>>> > After the eject, system comes up much more noisily / green, so
>>>> probably reassuring.
>>>> >
>>>> > However, it comes up under 4.9.59-0-*hardened or 4.9.65 vanilla, so
>>>> one can't use the ISO's 4.9.63 kernel to boot a rescue. If it isn't too
>>>> much trouble, for the final 3.7 release, it'd be good if the kernel matched
>>>> on the ISO's boot and the ISO's apk.
>>>> >
>>>> > Guest additions (mostly not needed by me) are hard to load, since
>>>> they are in /lib/modules/4.9.65-* but the kernel is actually 4.9.59-*. Once
>>>> that is fixed, all good.
>>>> >
>>>> >
>>>> > Standard, Extended, Vanilla EFI issue (vfat.fsck)
>>>> >
>>>> > * 1.5GB disk required
>>>> > * vfat.fsck
>>>> >
>>>> > Alpine is known for it's very small size. The ISOs are a nice size. I
>>>> have a VM that runs off a 10MB iso, and the Virtual ISO is good at under
>>>> 40MB. I was trying out ubuntu variations and had to choose my physical
>>>> network to handle the each-iso-is-a-gig downloads. Alpine is very easy to
>>>> try. However, the new EFI install creates a 500MB /boot/efi partition to
>>>> hold 500KB of boot data. It might be worthwhile to check the total size of
>>>> the disk, and not use so much on it for smaller (1 or 2 gig) drives.
>>>> >
>>>> > More importantly /boot/efi fails its fsck due to a syntax error:
>>>> >
>>>> > fsck.vfat: unrecognized option: C
>>>> >
>>>> > This is from /etc/init.d/fsck. Since the fsck service fails, most of
>>>> the rest of the boot is not really attempted. / is mounted read-only.
>>>> Network is down.
>>>> >
>>>> > Remounting / rw, removing the -C0 from fsck, and rebooting is OK. You
>>>> still get a scary error, but the system comes up fine.
>>>> >
>>>> > See also https://bugs.alpinelinux.org/issues/8090 for raspberry pi
>>>> breakage and suggested fix (seems reasonable to me: write a simple shell
>>>> script wrapper for fsck.vfat and fsck.fat that ignores ext2/3/4 arguments).
>>>> >
>>>> > I think this must be fixed before a release.
>>>> >
>>>> >
>>>> > Virtual EFI fail
>>>> >
>>>> > Virtual's virthardened kernel lacks EFI support (even in module
>>>> form). Also it is worse on VirtualBox EFI since it also lacks video
>>>> drivers. The serial console works, but is hard to use (telnet lacks
>>>> tab-completion, GNU screen requires socat, and only one terminal). Since
>>>> /sys/firmware/efi is not available, setup-disk does not realize it was an
>>>> EFI boot and does everything wrong.
>>>> >
>>>> > I suspect it would be a good idea to include EFI support for the
>>>> Virtual ISO / virthardened kernel too. The fix would make Alpine seem more
>>>> reliable and would cut down on the "why is the screen black?" questions.
>>>> However, to workaround it is easy to turn off EFI, and legacy BIOS is more
>>>> standard for virtual machines, I believe.
>>>> >
>>>> >
>>>> >
>>>> > ---
>>>> > Unsubscribe: alpine-devel+unsubscribe_at_lists.alpinelinux.org
>>>> > Help: alpine-devel+help_at_lists.alpinelinux.org
>>>> > ---
>>>> >
>>>>
>>>>
>>>>
>>>> ---
>>>> Unsubscribe: alpine-devel+unsubscribe_at_lists.alpinelinux.org
>>>> Help: alpine-devel+help_at_lists.alpinelinux.org
>>>> ---
>>>>
>>>>
>>>
>>>
>>
>
> --
-- Kiyoshi Aman



---
Unsubscribe:  alpine-devel+unsubscribe_at_lists.alpinelinux.org
Help:         alpine-devel+help_at_lists.alpinelinux.org
---
Received on Fri Dec 01 2017 - 16:49:24 GMT