Re: GRUB testers on SPARC needed

[Frank Scheiner <frank.scheiner@web.de>]

Hi Adrian,

On 02/27/2018 11:42 AM, John Paul Adrian Glaubitz wrote:
> On 02/26/2018 05:55 PM, Frank Scheiner wrote:
> > Thanks for the upload. The new version works like version 
> > 2.02-2+sparc64.3 for me.
> Frank, if you like, you can help improving grub-installer for sparc64.

Sure! :-) I had a first look and also did some testing.

I did use the image from 2018-02-16 for my testing though, as the newer 
one from 2018-02-27 didn't work for me. When selecting the menu option 
to load additional installer components I get a red screen 
("Installation step failed") and looking into the log I see the following:

```
[...]
Mar  2 20:34:21 cdrom-retriever: warning: File 
/cdrom/dists/sid/main/debian-installer/binary-sparc64/Packages does not 
exist.
Mar  2 20:34:21 main-menu[220]: (process:836): Segmentation fault 

Mar  2 20:34:21 kernel: [  151.303940] anna[838]: segfault at 0 ip 
ffff800100497648 (rpc ffff800100308f74) sp 000007feff91a961 error 1 in 
libc.so.6[ffff800100410000+15e000] 

Mar  2 20:34:21 main-menu[220]: WARNING **: Configuring 'load-cdrom' 
failed with error code 139
Mar  2 20:34:21 main-menu[220]: WARNING **: Menu item 'load-cdrom' failed.
```

As I don't have a T4 (=minimum requirement for GPT support according to 
Eric's wiki page) I cannot test booting from a GPT partitioned disk. But 
I assume I can still test GRUB installations to GPT partitioned disks. 
So I rewrote `/lib/partman/lib/disk-label.sh` to propose a default 
partitioning scheme depending on the existence of a properly named file 
in `/tmp`:

```
[...]
sparc|sparc64)
    if [ -e /tmp/gpt ]; then
        echo gpt
    elif [ -e /tmp/sun ]; then
        echo sun
    else
        echo UNKNOWN
    fi;;
[...]
```

...which also works around the missing new subarch hardware detection on 
the older image I had to use.

## Results ##

Up until now I got GRUB2 installations working automatically for disks 
with Sun disklabel and GPT partitioned disks. For GPT I had to manually 
partition the disk to get that required BIOS_GRUB partition. In the end 
it worked with the following layout:

```
# parted /dev/sda print
Model: ATA Samsung SSD 850 (scsi)
Disk /dev/sda: 250GB
Sector size (logical/physical): 512B/512B
Partition Table: gpt
Disk Flags:

Number  Start   End     Size    File system     Name  Flags
1       1049kB  7340kB  6291kB                        bios_grub
2       7340kB  263MB   256MB   ext2
3       263MB   240GB   240GB   ext4
4       240GB   250GB   9796MB  linux-swap(v1)
```

So we should also have a look into the recipe for GPT partitioning on 
sparc(64).

The modifications I made autoselect (1) the partition for `/boot` as 
target if a Sun disklabel is used and (2) the whole disk containing the 
partition for `/boot` as target if the disk is GPT partitioned. A user 
is never asked to select a target device.

Is that what we want?

> We have extended the hardware detection in debian-installer on sparc and
> sparc64 now, so that the installer can detect whether the hardware supports
> GPT partition tables or not.

Nice. I have to say I never noticed that the subarch was always 
"generic" for sparc(64) up until your changes ([1]).

[1]: 
https://anonscm.debian.org/cgit/d-i/libdebian-installer.git/commit/src/system/subarch-sparc-linux.c?id=aa1ce486176b0cb24f5783d67eb28f80128aafb3

> This information should be used here:
>
> > https://anonscm.debian.org/git/d-i/grub-installer.git/tree/grub-installer#n628 
>
> and here:
>
> > https://anonscm.debian.org/git/d-i/grub-installer.git/tree/grub-installer#n853 
>
> and here:
>
> > https://anonscm.debian.org/git/d-i/grub-installer.git/tree/grub-installer#n867 

I don't think we can rely on the subarch value in this case, but have to 
determine the actual partitioning scheme (available in `bootfslabel` 
when running `grub-installer` or via `partmap <DISK>`) to select the 
correct installation method. This because at least for sun4v with GPT 
support users could also manually select to use a Sun disklabel. I used 
the `bootfslabel` variable in my modifcations.

Up until now I tested with a single disk installed, but I'll also test 
with multiple disks installed in my T5220 to see if this will make any 
difference.

I can come up with a patch the next days perhaps. This should be much 
smaller than the patchset for NewWorld Power Macs.

Cheers,
Frank