[ipxe-devel] initrd corruption since commit ef038491858cb51f8aa17b1f6e50444d2e627413

[Christian Iwata Nilsson]

On Fri, 25 Apr 2025, 23:13 Sven Dreyer via ipxe-devel, <
ipxe-devel at lists.ipxe.org> wrote:

> Dear iPXE developers,
>
> after the easter holidays, I just noticed that I can neither boot my
> Debian amd64 nfsroot system nor my Windows 10 PE x86_64 any longer.
>
> The iPXE Debian nfsroot thing worked like a charm with various Debian
> releases on amd64 as well as on i386.
>
> Now it doesn't boot because the initrd seems corrupted:
> Initramfs unpacking failed: invalid magic at start of compressed archive
>
> Comparable problems apply to wimboot 2.8.0 trying to boot a Windows PE
> 10.0-26100 x86_64:
> Unexpected output length 0x0 (expected 0x8000)
> Could not patch WIM boot.wim
>
> Setup is:
> - x86_64 on BIOS (not UEFI)
> - PXE boot configured
> - Integrated PXE loads ipxe.pxe (self compiled from git master)
> - iPXE loads script
> - script for Debian nfsroot loads both initrd and kernel via nfs:// uri
> and boots kernel with append parameters
> - script for Windows PE loads wimboot 2.8.0 and further windows files
> over HTTP
>
> What I noticed is:
> - The problem occurs with ipxe.pxe, undionly.kpxe and undionly.kkpxe
> - When switching to UEFI (and using ipxe.efi), the problem does NOT occur
> - A git bisect reveals that ef038491858cb51f8aa17b1f6e50444d2e627413 is
> the first bad commit.
> - Before commit ef038491858cb51f8aa17b1f6e50444d2e627413, everything
> works for me.
>
> Is there anything I can do to track this problem down?
>
> Thanks and best regards,
> Sven
>

Add imgstat, and maybe a pause before boot, post that output, then add
imgfree before your initrd and note any difference from imgstat, and note
if it makes any difference to the boot process.

I would however not think that this would apply in the same way to wimboot,
and you should have the same issues in EFI, but again compare imgstat
output.

There is https://github.com/ipxe/ipxe/issues/1454 pointing to the same
commit, however there it seems to be other data before cpio archive
(initrd), which I wouldn't expect to work. So maybe there has actually been
a different bug before, hidden due something else changing how the data is
packed.

/Christian


> _______________________________________________
> ipxe-devel mailing list
> ipxe-devel at lists.ipxe.org
> https://lists.ipxe.org/mailman/listinfo/ipxe-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ipxe.org/pipermail/ipxe-devel/attachments/20250426/08ab45a8/attachment.htm>