Web lists-archives.com

Re: Installing Debian on a Minnowboard Turbot with installer on USB stick?




Hi Thomas,

On Jun 3, 2018, at 1:26 AM, Thomas Schmitt <scdbackup@xxxxxxx> wrote:

> Hi,
> 
> Rick Thomas wrote:
>> Instead, I used
>>   firmware-9.4.0-amd64-netinst.iso
>> to avoid any possible alpha/testing anomalies.
> 
> Normally i'd say that there is no decisive difference to expect.
> In any case copying a "netinst" ISO is much faster than a "DVD-1".
> 
> 
>> When I tried part (a) — just a simple dd — with
>> debian-9.4.0-amd64-netinst.iso, the minnowboard immediately recognized it as
>> bootable and presented it as part of the EFI menu labeled “EFI USB Device”.
> 
> Normally this difference should not appear. Very strange.
> 
> But we can now classify the demand for GPT on
>  https://minnowboard.org/tutorials/best-practice-boot-media-selection
> as spin-off of the usual rumor that EFI needs GPT.
> 
> 
>> it booted without incident and ran the installer.
>> When it came to the part where it looks for a CD, it easily found the USB
>> stick
> 
> This was expectable. As long as the ISO is marked as partition, the
> script in the initrd should find the device.
> 
> 
>> tried
>> part (a) with “firmware-buster-DI-alpha2-amd64-DVD-1.iso”.  And guess what!
>> That worked too, just the same as the other two.
> 
> Hrmpf. At least above "normally"s are not generally put in doubt.
> It would be another riddle if one ISO would fail reproducibly while
> the other succedds reproducibly.
> 
> Maybe you changed a setting in the firmware ?

In the previous experiments (where I modified the partition tables on the stick after dd-ing) I did try some different setting in the firmware.  But for the above described experiments, I reset the firmware to factory defaults before starting the process and made no changes thereafter.


>> or there’s something flakey about the USB ports on this device
> 
> But why did the manual start of “fs0:\efi\boot\bootx64.efi” ran the
> boot loader, loaded the kernel, and came far enough to run into the
> software shortcoming of device detection ?

I’m not absolutely sure that the manual start was necessary.  I may have gotten used to never seeing the “EFI USB Device” menu entry.  So when the stars aligned and the stick *was* seen, I didn’t notice.  So, maybe, the fact that I got it to work from the command line was just because it would have worked anyway that time from the EFI menu.

>> In any case, I apologize for all the noise.
> 
> I doubt that this was a hallucination.
> If not the hardware (stick and port) makes itself a suspect by failing
> in normal operation, then i’d bet on the firmware's settings.

I’m pretty sure that everything I’ve been seeing can be explained if we assume the minnowboard firmware has trouble recognizing a USB3.0 device as bootable.


> -------------------------------------------------------------------------
> 
>> Thomas, If you would like to follow up on the inability of the installer to
>> recognize itself as a suitable installation CD
> 
> I believe to understand that problem. Let's see whether debian-cd will
> find a time slot to discuss it and whether trying the devices found by
> "list-devices disk" (/dev/sda, /dev/sdb, /dev/sdc on my machine) is
> harmless enough.
> 
> It would be interesting to study the history of cdrom-detect.postinst
> (package "cdrom-detect", 1600+ commits in
> https://salsa.debian.org/installer-team/cdrom-detect/commits/master).
> cdrom-detect/blame brings me to some forth and back, which probably
> does not tell the reason for the decision not to look for "disk".
> 
> I'd need a comfortable method to look at the revison _before_ a commit
> in order to get the next older blaming.
> Or a method to see only commits which affect the interesting code parts
>   devices="$(list-devices cd; list-devices maybe-usb-floppy)"
> and
>   devices="$(list-devices usb-partition)"
>   ...
>        if try_mount $device $CDFS; then
> 
> Any GitLab experts here who could help me navigate ?
> Or git experts who can augment my knowledge beyond "clone", "add",
> "commit", and "push", so that i can inspect a clone locally ?
> 
> 
> Have a nice day :)
> 
> Thomas
>