[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Debian Bookworm RC 1 installer- a Bug?



On Sun 16 Apr 2023 at 07:19:18 (-0700), Peter Ehlert wrote:
> On 4/9/23 08:57, David Wright wrote:
> > On Wed 05 Apr 2023 at 07:03:41 (-0700), Peter Ehlert wrote:
> > > Debian Bookworm RC 1 installer
> > > Damned nice, the improvements are appreciated.
> > I ran rc1 in my usual manner, and the only difference I noticed was
> > the one extra question about non-free firmware, to which I replied
> > yes. (There may well be improvements under the hood, so to speak.)
> > Oh, and the initrd is somewhat larger, as per usual.
> > 
> > > using the new debian-bookworm-DI-rc1-amd64-netinst.iso
> > > Legacy install, GPT partition
> > I assume Legacy means BIOS booting. Same here, but only one disk.
> correct. different term, same thing. Not UEFI
> > 
> > > graphic install, manual partitioning
> > > Mate Desktop (others were deselected)
> > Non-graphical here, a suitable partition existed, and only
> > standard and SSH server software was installed.
> > 
> > > WiFi firmware:
> > Untested as this machine is a 2006-vintage mini-tower lacking wifi.
> > 
> > [ snipped narrative of later network-switching ]
> > 
> > > Boot Loader:
> > > all disk drives were detected, however the one with the bios_grub
> > > partition was highlighted
> > I can't recall seeing anything other than the first item highlighted,
> > ie "Enter device manually", at least with the non-graphical installer
> > in expert mode. I selected the (sole) hard drive, item 2. The only
> > remaining item was the USB stick containing the installer ISO.
> > 
> > As expected nowadays, when the machine rebooted, the Grub menu
> > had only two lines, both pointing to the newly installed system.
> > (I hadn't made any attempt to counteract GRUB_DISABLE_OS_PROBER
> > during my installation.) So Grub was correctly installed in the
> > MBR, and the rest of Grub occupied d400 bytes of /dev/sda1 (the
> > 3MB BIOS boot partition on the single disk).
> > 
> > > =============
> > > second try, using the debian-live-bkworm-DI-rc1-amd64-mate.iso
> > > same machine and again Legacy install, GPT partition
> > > however I did NOT install from the live session:
> > > I chose to go directly to install rather than the Calamares installer
> > > then manual partitioning
> > > 
> > > Boot Loader:
> > > all drives were detected, however the one with the bios_grub partition
> > > was NOT highlighted, but I did select it.
> > > GRUB was Not properly installed, my former grub menu was still active.
> > How did you determine that it was the previous menu. Wouldn't it look
> > just the same?
> I enable GRUB_DISABLE_OS_PROBER so that the various other operating
> systems are shown
> if the new GRUB is properly installed I get the "new" one item only
> GRUB display.
> then when I boot the new OS I again enable GRUB_DISABLE_OS_PROBER and
> update GRUB
> > 
> > > *** I tried a second time, same as above being super careful, same result.
> > > 
> > > I then booted with my default system, ran grub-install /dev/sde &&
> > > update-grub
> > > then "new" system was on my boot menu.
> > > then booted and it ran as expected.

So you installed Grub on /dev/sde.

> > Which method did you use to boot the "default" system (which I assume
> > is bullseye, in a different partition on one or other of the disks),
> > in view of the rather sparse menu from grub.cfg on the new system?
> I boot with the "old" GRUB menu as explained above...it has Several
> operating systems listed, my old default OS is still at the top of the
> list.
> > 
> > > back to the WiFi dongle, again the obscure firmware was properly installed
> > > 
> > > Is this a Bug or a user/hardware issue?
> > Presumably we are now back to talking about Grub.
> > 
> > If you still have access to the bookworm system, you can check whether
> > it claimed to have completed installing Grub successfully. You should
> > see lines like:
> > 
> >    grub-installer: info: Installing grub on '/dev/sda'
> >    grub-installer: info: grub-install does not support --no-floppy
> >    grub-installer: info: Running chroot /target grub-install  --force "/dev/sda"
> >    grub-installer: Installing for i386-pc platform.
> >    grub-installer: Installation finished. No error reported.
> >    grub-installer: info: grub-install ran successfully
> > 
> > in /var/log/installer/syslog.
> Thanks, I did not know where to look or what to look for.

I've tidied these lines:

> ===================
> Apr  5 12:59:44 grub-installer: info: Identified partition label for
> /dev/sdb12: gpt

> Apr  5 13:01:03 grub-installer: info: Installing grub on '/dev/sdb'
> Apr  5 13:01:03 grub-installer: info: grub-install does not support
> --no-floppy
> Apr  5 13:01:03 grub-installer: info: Running chroot /target
> grub-install  --force "/dev/sdb"
> Apr  5 13:01:03 grub-installer: Installing for i386-pc platform.
> Apr  5 13:01:13 grub-installer: grub-install: warning: this GPT
> partition label contains no BIOS Boot Partition; embedding won't be
> possible.
> Apr  5 13:01:13 grub-installer: grub-install: warning: Embedding is
> not possible.  GRUB can only be installed in this setup by using
> blocklists.  However, blocklists are UNRELIABLE and their use is
> discouraged..
> Apr  5 13:01:13 grub-installer: Installation finished. No error reported.
> Apr  5 13:01:13 grub-installer: info: grub-install ran successfully
> 
> ====

So that was installing Grub on /dev/sdb. I'm confused. This is a BIOS
machine, so it boots via the MBR. Then it's meant to use blocklists?

But you installed Grub on /dev/sde. Is that using blocklists? Are you
always booting from the same MBR?

Without tying all this down 100%, you have no bug IMO. It seems you
have an extremely fragile boot implementation.

> New Information:
> in my effort to eliminate the hardware possibility I made several new
> installs on three other machines, all with multiple drives, and only
> one with the bios_grub partition
> 
> at the close of the install we are presented with a "Install the GRUB
> boot loader" window/menu
> it asks "install the GRUB boot loader to your primary drive" with Yes
> and No options
> 
> * if I select No it presents a list of my physical drives...
> choosing the one with the bios_grub partition does not work, GRUB is
> not installed and the same fails are shown in the log.

You're outside my comfort zone. My understanding is that you don't
refresh the MBR, and write a blocklist. (I didn't see failure in
the log, just a robust warning.) But I don't know where that blocklist
is (I don't use them). It would be very easy for them to be overwritten.
Perhaps boot-init-script would help here.

> * if I select Yes it again presents the same list of my physical drives...
> choosing the one with the bios_grub partition Does work, GRUB is
> installed as expected

I though you wrote earlier that that didn't work. (Nine lines after
 "second try, using the debian-live-bkworm-DI-rc1-amd64-mate.iso"
including blank lines.)

> logically, I select NO because I have no clue which is the Primary
> Drive and fully understand that it needs to point to the one with the
> bios_grub partition.

What "needs to point to"? AIUI the Primary drive is the one the BIOS
reads the MBR from by default, ie if you haven't overridden it with
the one-shot boot device menu (ie what Dell typically put on F12).
So the BIOS has to point to the MBR which then points Grub at the
BIOS Boot Partition, when there is one (or more), and that gets you
to a Grub grub.cfg menu, and an OS.

So I'm not confident where your active MBR is (sda, or sdb, or sde;
each disk can have an MBR), and why Grub is writing blocklists
(implying it doesn't know where the BIOS Boot Partition is).
So I'd say the last paragraph, quoted below, still applies.

There's a fair amount of information on the web, too, about
dual-booting Grub, which might be worth perusing.

> SO: where do I report this Bug/Anomaly?
> I assume it should go to the debian-boot list

I would say it's rather premature as a bug report, but more
experienced eyes might help with what's going wrong where.

> > You could install and run boot-info-script, which provides details of
> > how the system boots, particularly where the MBR code looks for the
> > BIOS boot partition (ie core.img). BTW do any other disks in this
> > machine have BIOS boot partitions? (I've one on all my internal disks.)
> thanks for that thought
> > 
> > But as far as we're concerned, I think more information is needed,
> > like what disks there are on the system, which disk the BIOS is
> > reading the MBR from, the final listing from the partitioner,
> > particularly any BIOS boot partitions, and so on. Without all that
> > in the narrative, there's no telling whether it's a bug or not.

Cheers,
David.


Reply to: