Can't boot kernel 6.17.4+ via rEFInd

Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi All,

I've got a problem booting the stable kernel 6.17.4 and 6.17.5 after
updating from 6.17.3 via make oldconfig. Out of my 4 amd64 machines
running Gentoo, two are unable to boot with rEFInd showing "Error: Not
Found while loading <kernel file name>." Bisecting kernel sources
points to commit 5b5cdb1fe434e8adc97d5037e6d05dd386c4c4c6.

All four machines use very similar toolchains: GCC 15.2.0 and
15.2.1_p20251018, binutils 2.45.

https://bugzilla.kernel.org/show_bug.cgi?id=220700 has kernel config attached.

Would appreciate any hints or suggestions.

Best regards,
Eugene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Nathan Chancellor]

Hi Eugene,

On Sun, Oct 26, 2025 at 09:25:45AM +0100, Eugene Shalygin wrote:
> I've got a problem booting the stable kernel 6.17.4 and 6.17.5 after
> updating from 6.17.3 via make oldconfig. Out of my 4 amd64 machines
> running Gentoo, two are unable to boot with rEFInd showing "Error: Not
> Found while loading <kernel file name>." Bisecting kernel sources
> points to commit 5b5cdb1fe434e8adc97d5037e6d05dd386c4c4c6.
> 
> All four machines use very similar toolchains: GCC 15.2.0 and
> 15.2.1_p20251018, binutils 2.45.
> 
> https://bugzilla.kernel.org/show_bug.cgi?id=220700 has kernel config attached.
> 
> Would appreciate any hints or suggestions.

Does applying

  https://lore.kernel.org/all/20251026202100.679989-1-dimitri.ledkov@surgut.co.uk/

help with this issue? Also, does 6.18-rc2 (or -rc3 if it is out by the
time you see this) have the same issue?

Cheers,
Nathan

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi Nathan,

On Sun, 26 Oct 2025 at 22:13, Nathan Chancellor <nathan@kernel.org> wrote:
> Does applying
>
>   https://lore.kernel.org/all/20251026202100.679989-1-dimitri.ledkov@surgut.co.uk/
>
> help with this issue? Also, does 6.18-rc2 (or -rc3 if it is out by the
> time you see this) have the same issue?

Thanks for looking into this!

Neither the patch nor 6.18-rc3 work. I'm going to try to revert
5b5cdb1fe434e8adc97d5037e6d05dd386c4c4c6 together with anything that
is needed to apply the reversing patch.

Cheers,
Eguene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi Nathan,

On Mon, 27 Oct 2025 at 20:35, Eugene Shalygin <eugene.shalygin@gmail.com> wrote:
> Neither the patch nor 6.18-rc3 work. I'm going to try to revert
> 5b5cdb1fe434e8adc97d5037e6d05dd386c4c4c6 together with anything that
> is needed to apply the reversing patch.

Replacing CONFIG_KERNEL_ZSTD with  CONFIG_KERNEL_GZIP made the kernel
bootable. What does that mean?

Cheers,
Eugene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Nathan Chancellor]

On Mon, Oct 27, 2025 at 09:41:25PM +0100, Eugene Shalygin wrote:
> Hi Nathan,
> 
> On Mon, 27 Oct 2025 at 20:35, Eugene Shalygin <eugene.shalygin@gmail.com> wrote:
> > Neither the patch nor 6.18-rc3 work. I'm going to try to revert
> > 5b5cdb1fe434e8adc97d5037e6d05dd386c4c4c6 together with anything that
> > is needed to apply the reversing patch.
> 
> Replacing CONFIG_KERNEL_ZSTD with  CONFIG_KERNEL_GZIP made the kernel
> bootable. What does that mean?

Hmmmm, I am not sure... That seems rather odd within the context of the
flagged change.

Could you post the output of 'readelf -S vmlinux vmlinux.unstripped'
from the broken and good builds? Does rEFInd have any sort of additional
debugging to see why/what it is complaining about not being able to
find?

Cheers,
Nathan

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

[-- Attachment #1: Type: text/plain, Size: 824 bytes --]

Hi Nathan,

On Tue, 28 Oct 2025 at 18:45, Nathan Chancellor <nathan@kernel.org> wrote:
> > Replacing CONFIG_KERNEL_ZSTD with  CONFIG_KERNEL_GZIP made the kernel
> > bootable. What does that mean?
>
> Hmmmm, I am not sure... That seems rather odd within the context of the
> flagged change.

Huh! The fourth of my machines with the similar config (also fully
updated ~amd64 Gentoo) with rEFind boots kernel 6.17.5 with
CONFIG_KERNEL_ZSTD enabled.

> Could you post the output of 'readelf -S vmlinux vmlinux.unstripped'
> from the broken and good builds?

Attached.

> Does rEFInd have any sort of additional
> debugging to see why/what it is complaining about not being able to
> find?

No, unfortunately. However, I was unable to find the first part of the
error message ("Not Found") in the rEFInd sources.

Cheers,
Eugene

[-- Attachment #2: readelf-vmlinux.tar.zst --]
[-- Type: application/zstd, Size: 1911 bytes --]

Re: Can't boot kernel 6.17.4+ via rEFInd

[Nathan Chancellor]

Hi Eugene,

On Wed, Oct 29, 2025 at 09:07:20PM +0100, Eugene Shalygin wrote:
> On Tue, 28 Oct 2025 at 18:45, Nathan Chancellor <nathan@kernel.org> wrote:
> > > Replacing CONFIG_KERNEL_ZSTD with  CONFIG_KERNEL_GZIP made the kernel
> > > bootable. What does that mean?
> >
> > Hmmmm, I am not sure... That seems rather odd within the context of the
> > flagged change.
> 
> Huh! The fourth of my machines with the similar config (also fully
> updated ~amd64 Gentoo) with rEFind boots kernel 6.17.5 with
> CONFIG_KERNEL_ZSTD enabled.
> 
> > Could you post the output of 'readelf -S vmlinux vmlinux.unstripped'
> > from the broken and good builds?
> 
> Attached.

Unfortunately, nothing appears to be out of the ordinary there... Only
.modinfo gets removed from the vmlinux.unstripped to vmlinux stage.

> > Does rEFInd have any sort of additional
> > debugging to see why/what it is complaining about not being able to
> > find?
> 
> No, unfortunately. However, I was unable to find the first part of the
> error message ("Not Found") in the rEFInd sources.

I am guessing the "Not Found" is EFI_NOT_FOUND, so I think this is
coming from StartEFIImage() in refind/launch_efi.c:

  https://sourceforge.net/p/refind/code/ci/253abe5c4af58d044912517daa567c4440612c46/tree/refind/launch_efi.c#l290

I am not sure where the LoadImage call comes from (the firmware?) so I
don't really know why that call would fail... Has CONFIG_KERNEL_ZSTD
always worked in this configuration? Presumably since you are not seeing
"Invalid loader file", you are going into the

  if ((LoaderType == LOADER_TYPE_EFI) || (LoaderType == LOADER_TYPE_GZIP)) {

branch there. I am at a bit of a loss here. Ard, could you see any
obvious reason from an EFI perspective why

  3e86e4d74c04 ("kbuild: keep .modinfo section in vmlinux.unstripped")

would cause issues with booting via rEFInd? You can see Eugene's
original report at

  https://lore.kernel.org/CAB95QARfqSUNJCCgyPcTPu0-hk10e-sOVVMrnpKd6OdV_PHrGA@mail.gmail.com/

and subsequent replies in case they are relevant.

Cheers,
Nathan

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi Nathan,

On Fri, 31 Oct 2025 at 23:07, Nathan Chancellor <nathan@kernel.org> wrote:
> Unfortunately, nothing appears to be out of the ordinary there... Only
> .modinfo gets removed from the vmlinux.unstripped to vmlinux stage.

I've made one more experiment: kernel 6.17.5, built on a machine which
shows no problem, using the .config file  from the machine where it
does not boot, produces binaries that boot at the machine with the
problem. These two machines are nearly identical in terms of installed
software packages and their versions.

Now, can I conclude that it is the toolchain which is faulty on two
machines? And in which way?

Cheers,
Eugene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Ard Biesheuvel]

On Fri, 31 Oct 2025 at 23:07, Nathan Chancellor <nathan@kernel.org> wrote:
>
> Hi Eugene,
>
> On Wed, Oct 29, 2025 at 09:07:20PM +0100, Eugene Shalygin wrote:
> > On Tue, 28 Oct 2025 at 18:45, Nathan Chancellor <nathan@kernel.org> wrote:
> > > > Replacing CONFIG_KERNEL_ZSTD with  CONFIG_KERNEL_GZIP made the kernel
> > > > bootable. What does that mean?
> > >
> > > Hmmmm, I am not sure... That seems rather odd within the context of the
> > > flagged change.
> >
> > Huh! The fourth of my machines with the similar config (also fully
> > updated ~amd64 Gentoo) with rEFind boots kernel 6.17.5 with
> > CONFIG_KERNEL_ZSTD enabled.
> >
> > > Could you post the output of 'readelf -S vmlinux vmlinux.unstripped'
> > > from the broken and good builds?
> >
> > Attached.
>
> Unfortunately, nothing appears to be out of the ordinary there... Only
> .modinfo gets removed from the vmlinux.unstripped to vmlinux stage.
>
> > > Does rEFInd have any sort of additional
> > > debugging to see why/what it is complaining about not being able to
> > > find?
> >
> > No, unfortunately. However, I was unable to find the first part of the
> > error message ("Not Found") in the rEFInd sources.
>
> I am guessing the "Not Found" is EFI_NOT_FOUND, so I think this is
> coming from StartEFIImage() in refind/launch_efi.c:
>
>   https://sourceforge.net/p/refind/code/ci/253abe5c4af58d044912517daa567c4440612c46/tree/refind/launch_efi.c#l290
>
> I am not sure where the LoadImage call comes from (the firmware?) so I
> don't really know why that call would fail... Has CONFIG_KERNEL_ZSTD
> always worked in this configuration? Presumably since you are not seeing
> "Invalid loader file", you are going into the
>
>   if ((LoaderType == LOADER_TYPE_EFI) || (LoaderType == LOADER_TYPE_GZIP)) {
>
> branch there. I am at a bit of a loss here. Ard, could you see any
> obvious reason from an EFI perspective why
>
>   3e86e4d74c04 ("kbuild: keep .modinfo section in vmlinux.unstripped")
>
> would cause issues with booting via rEFInd?

For some reason, this change seems to result in a PE/COFF header
change that results in the EFI firmware refusing to load it.

Eugene, can you try whether the image can be loaded by the EFI shell
directly? You may have to rename the file and give it a .efi
extension, but otherwise, you should be able to boot it using

Shell> fs0:bzImage.efi root=/dev/....   etc etc

(or wherever the file resides)

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi Ard and Nathan,

On Mon, 3 Nov 2025 at 16:00, Ard Biesheuvel <ardb@kernel.org> wrote:
> Eugene, can you try whether the image can be loaded by the EFI shell
> directly? You may have to rename the file and give it a .efi
> extension, but otherwise, you should be able to boot it using

Did that with the kernel file and it loaded up until the non syncing
VFS error, but I suppose that is enough to state that the firmware can
boot the kernel image which rEFInd is unable to, and I should look for
answers with rEFInd author?

Cheers,
Eugene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Ard Biesheuvel]

On Mon, 3 Nov 2025 at 22:22, Eugene Shalygin <eugene.shalygin@gmail.com> wrote:
>
> Hi Ard and Nathan,
>
> On Mon, 3 Nov 2025 at 16:00, Ard Biesheuvel <ardb@kernel.org> wrote:
> > Eugene, can you try whether the image can be loaded by the EFI shell
> > directly? You may have to rename the file and give it a .efi
> > extension, but otherwise, you should be able to boot it using
>
> Did that with the kernel file and it loaded up until the non syncing
> VFS error, but I suppose that is enough to state that the firmware can
> boot the kernel image which rEFInd is unable to, and I should look for
> answers with rEFInd author?
>

Indeed. Not saying there isn't anything to fix on the kernel side, but
narrowing it down needs to be done on the side of the loader, it
seems.

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

On Mon, 3 Nov 2025 at 22:28, Ard Biesheuvel <ardb@kernel.org> wrote:
> Indeed. Not saying there isn't anything to fix on the kernel side, but
> narrowing it down needs to be done on the side of the loader, it
> seems.

Thanks a lot!

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

I've got no response from the rEFInd author yet, but another user
reported (https://bugzilla.kernel.org/show_bug.cgi?id=220731) a very
similar problem with GRUB and ZSTD compression.

Cheers,
Eugene

Re: Can't boot kernel 6.17.4+ via rEFInd

[Nathan Chancellor]

Hi Eugene,

On Sat, Nov 08, 2025 at 10:06:20AM +0100, Eugene Shalygin wrote:
> I've got no response from the rEFInd author yet, but another user
> reported (https://bugzilla.kernel.org/show_bug.cgi?id=220731) a very
> similar problem with GRUB and ZSTD compression.

Small necro bump: I think this might be fixed in mainline now with
commit 8678591b4746 ("kbuild: Split .modinfo out from ELF_DETAILS"). It
should apply relatively cleanly to stable versions (and should make its
way there in due time due to the stable tag).

Cheers,
Nathan

Re: Can't boot kernel 6.17.4+ via rEFInd

[Eugene Shalygin]

Hi Nathan,

On Sat, 7 Mar 2026 at 09:08, Nathan Chancellor <nathan@kernel.org> wrote:
> Small necro bump: I think this might be fixed in mainline now with
> commit 8678591b4746 ("kbuild: Split .modinfo out from ELF_DETAILS"). It
> should apply relatively cleanly to stable versions (and should make its
> way there in due time due to the stable tag).

Thank you, I will try rEFInd again, although the project looks abandoned.

Cheers,
Eugene