"Invalid signature" issue on dev kernel launch

[Valdis Kl=?utf-8?Q?=c4=93?=tnieks]

On Tue, 23 Mar 2021 22:36:33 -0400, Aruna Hewapathirane said:

> I also see:
> CC      drivers/cpufreq/cpufreq_ondemand.o
> drivers/cpufreq/cpufreq_ondemand.c: In function ‘od_set_powersave_bias’:
> drivers/cpufreq/cpufreq_ondemand.c:446:1: warning: the frame size of 1032
> bytes is larger than 1024 bytes [-Wframe-larger-than=]
>   446 | }
>       | ^
> This is what causes the compile time errors with possible missing firmware

No it's not. That's just a *warning* and doesn't stop the build.  The reason
that we have that warning is because on most architectures, the entire kernel
stack has to fit into 2 contiguous 4k pages.

And a perfectly legal stack can be something like TCP delivering an NFS packet
which then goes to the NFS code, which then hands it off to the VFS layer,
which checks the page cache, which hands it off to XFS, which schedules I/O to
the filesystem, which gets handed to the block subsystem, which realizes that
the target partition is an LVM partition, and LVM realizes that the "physical
volume" is actually a dm-crypt, and then dm-crypt does some cryptography and
calls the block subsystem to write to the physical disk. Feel free to check the code
path yourself... :)

And all of that has to fit in 8K of stack. That's why we warn when things have
a large stack frame entry.

Anyhow, "possible missing firmware" isn't even at compile time - it's during
the build of the initramfs at *install* time.  And the most likely cause of
that is that missing external firmware that's not GPL - it's all confined to 2
graphics drivers - amdgpu and i915, so not at all related to a warning against
cpufreq...

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20210324/b31b353d/attachment.sig>