Cold boot consistently fails to load file from SD

[Mauro Condarelli]

Hi Daniel,
comments inline below.

On 9/19/20 9:15 PM, Daniel Schwierzeck wrote:
> Hi Mauro,
>
> Am Samstag, den 19.09.2020, 15:39 +0200 schrieb Mauro Condarelli:
>
>> Hi,
>> I'm facing a new problem for my Mt7628/vocore2 target.
>>
>> I moved the Linux kernel from a ext4 partition to the "proper" SquashFS partition (I say "proper" because I'm using a dual-system with fallback to avoid updating the currently working rootFS and kernel lies into rootFS).
>>
>> I am using u-boot v2020.10-rc3 plus the attached patches (one pertains SquashFS).
>>
>> Problem is a real cold boot (target off for more than one minute) leads to a load failure for the kernel. A warm boot ("reboot" from a running Linux) or a not-so-cold boot (turn target off and power it on again after a few seconds) seems to work as expected.
> which distro or Linux version do you use? Can you provide a full boot
> log?
I am using a self-compiled Buildroot RootFS, but this is hardly a problem.
Error is in pain u-boot.
I try to load Linux kernel in memory (from SD, from SquashFS) and this fails.
Linux has no chance to start at all.

> Linux MIPS had a bug with setting up memory for its exception vectors
> and handlers if the CPU didn't report support for vectored interrupts.
> mtmips is one of those CPUs. AFAIK this should have been fixed in
> mainline since some months. Could be one reason. 
I am using v2020.10-rc3 which should be "recent enough"... or am I wrong?

> You could also try to disable CONFIG_RESTORE_EXCEPTION_VECTOR_BASE in
> U-Boot. 
>
> Stefan also had cache issues on the Gardena board. He solved it with
> some memory allocation and copying, see arch/mips/mach-mtmips/cpu.c.
>
> The cmd_cache interface is not supported on MIPS as there is no
> reliable way to disable or enable caches or to separately control i-
> cache and d-cache.
>
> Can you send patches 0001 and 0002 to the mailing list in the proper
> way so that patchwork can pick them up? I'd like to prepare a small
> bugfix pull request for v2020.10. Thanks.
I'll do that tomorrow 'couse I'm too groggy now to be reliable ;)

Thanks
Mauro