Re: [PATCH 1/2] host/rootfs: use initramfs in "make run"

[Alyssa Ross <hi@alyssa.is>]

[-- Attachment #1: Type: text/plain, Size: 3305 bytes --]

José Pekkarinen <jose.pekkarinen@unikie.com> writes:

> On Thu, Sep 8, 2022 at 1:52 PM Alyssa Ross <hi@alyssa.is> wrote:
>
>> José Pekkarinen <jose.pekkarinen@unikie.com> writes:
>>
>> > On Thu, Sep 1, 2022 at 1:47 PM Alyssa Ross <hi@alyssa.is> wrote:
>> >
>> >> This will allow us to stop compiling e.g. the virtio-blk module into
>> >> the kernel, because it will be loaded by the initramfs.
>> >>
>> >> This introduces some duplication between the rootfs and initramfs's
>> >> Makefiles.  I don't think it's worth the effort at the moment to try
>> >> to reduce that, because it would come at the expense of additional
>> >> complexity in the Makefiles.  We can revisit this later if we want to.
>> >>
>> >> Signed-off-by: Alyssa Ross <hi@alyssa.is>
>> >
>> > Hi,
>> >
>> > This patchset introduces errors in the default qemu
>> >
>> > configuration of spectrum, where it is possible to see in the
>> > console log attempts to load broken aliases like:
>> >
>> > modprobe: FATAL: Module acpi:PNP0C0F: not found in directory
>> > /lib/modules/5.18.0
>>
>> This is pretty much expected.  The only job of the initramfs is to get
>> the root filesystem mounted, so most kernel modules are not included in
>> it to keep the size down.  It just contains modules related to block
>> devices, dm-verity, ext4, etc.  If a we try to load a driver in
>> initramfs, and it isn't available, that's fine, because when we get to
>> the root filesystem, we run mdevd-coldplug again, and the module will be
>> loaded at that point.
>>
>> The reason this didn't happen before when testing Spectrum in a VM is
>> that since all the drivers we needed were built in to the kernel, the
>> the block device nodes would be available before userspace even started,
>> meaning that the initramfs simply wouldn't have any time to try and fail
>> loading any other drivers before the rootfs was ready.  On hardware,
>> where the appropriate drivers wouldn't have been built in, I assume
>> these messages would already have been normal without this change,
>> although I didn't test.
>>
>> > As well as valid modules load failures like:
>> >
>> > modprobe: ERROR: could not insert 'vfio_pci': Invalid argument
>>
>> I wasn't able to reproduce this, with this series applied on top of
>> commit c0b9dff8653b59f5d2a24bb539cba6c91d3f7506,
>> Nixpkgs commit 4d05083dd894b73941e2e7d3b4f428b0ce1c7007, running
>> Spectrum with `make run' in the host/rootfs directory.  I started netvm
>> and verified that the QEMU ethernet device was successfully passed
>> through as well, so vfio-pci was definitely working.
>>
>> Does that differ to the versions you were using to test?  If so, could
>> you try with those versions and let me know if that works for you?
>>
>
> Yes, I was living in b01594b2c089ce2434dacddccf9a285af7334d24,
>
> right version on nixpkgs. Rebasing on upstream main solved the issue.
>
> Thanks!

Awesome!  If it's now working for you, want to send me some Tested-by:
lines for the patches you've tested, so they can be recorded in the
commit log?  (Just reply to the patches you've tested with "Tested-by:
José Pekkarinen <jose.pekkarinen@unikie.com>" on a line on its own, and
then it'll be automatically picked up when I apply the patches.)

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]