Re: Partition layout of ARM SD card images

On Mon, Jul 11, 2022 at 8:31 AM Rodney W. Grimes <
freebsd-rwg@gndrsh.dnsmgr.net> wrote:

> >
> > On 2022-Jul-10, at 14:34, Dr. Rolf Jansen <freebsd-rj@cyclaero.com>
> wrote:
> >
> > >> Am 10.07.2022 um 17:48 schrieb Mark Millard <marklmi@yahoo.com>:
> > >>
> > >> On 2022-Jul-10, at 12:26, Dr. Rolf Jansen <freebsd-rj@cyclaero.com>
> wrote:
> > >>
> > >>> For example let's have a llok on the partition layout of,
> FreeBSD-13.1-RELEASE-arm64-aarch64-RPI.img (the others are similar):
> > >>>
> > >>> # mdconfig -a -u 0 -t vnode -f
> diskimg/FreeBSD-13.1-RELEASE-arm64-aarch64-RPI.img
> > >>> # gpart show md0 md0s2
> > >>>
> > >>> =>     63  6291393  md0  MBR  (3.0G)
> > >>>        63     2016       - free -  (1.0M)
> > >>>      2079   102312    1  fat32lba  [active]  (50M)
> > >>>    104391  6187041    2  freebsd  (3.0G)
> > >>>   6291432       24       - free -  (12K)
> > >>>
> > >>> =>      0  6187041  md0s2  BSD  (3.0G)
> > >>>         0       57         - free -  (29K)
> > >>>        57  6186880      1  freebsd-ufs  (2.9G)
> > >>>   6186937      104         - free -  (52K)
> > >>>
> > >>> The start of the fat32 boot slice s1 (containing the u-boot) stuff
> is neither aligned to 1M nor to 4k, it starts on an odd base. The start of
> the BSD payload slice s2 and its size are odd as well. The padding of 57
> blocks within s2 lets the UFS partition start on a globally even base,
> namely 104391+57 = 104448, which as a matter of fact is 4k aligned
> (104448*512/4096 = 13056) and 1M aligned as well (104448*512/1024/1024 =
> 51), however all this keeps looking strange.
> > >>>
> > >>> Are there reasons for this partition layout besides making it look
> more interesting? If yes, some insights would be good.
> > >>
> > >> The layout details are more specific to the aarch64 RPi* context
> > >> than to general aarch64 SD card images. For example, the Rock64
> > >> image is different:
> > >>
> > >> # mdconfig -a -u 0 -t vnode -f
> FreeBSD-14.0-CURRENT-arm64-aarch64-ROCK64-20220708-a0b956f5ac5-256605.img
> > >> # gpart show md0
> > >> =>     40  6291376  md0  GPT  (3.0G)
> > >>      40    32728       - free -  (16M)
> > >>   32768   102400    1  efi  (50M)
> > >>  135168  6156160    2  freebsd-ufs  (2.9G)
> > >> 6291328       88       - free -  (44K)
> > >
> > > This is a GPT table, while the others are still MBR. Images which come
> with u-boot must have a different layout.
> >
> > I know it is a GPT table. That is part of the point about
> > the variety of contexts that there are across the Small
> > Board Computers.
> >
> > No SBC that has a U-Boot/whatever needing more space
> > than is provided below is going to use the same
> > 2079 figure:
> >
> > =>     63  ???  md0  ???  (?)
> >        63     2016       - free -  (1.0M)
>
> I read this thread.. and it keeps nagging me in
> the back of the head, I know this 2016 number.
> It is common when the sectors per track of a
> drive is reported as "32", its an attempt to
> 1M align such a drive, is somehow the image
> creation code picking up 32 to do the align
> calculation wrongly on a 63 sector/track
> image?
>
> >      2079   ??????    1  fat32lba  [active]  (?)
>
> The OP is correct, this is a horrid state of alignment
> and the cause should be tracked down and fixed!  I
> can see no valid reason to have an approx
> 1MB free hole that causes this missalignment,
> that free hole should be (2048-63)=1985
>

Yes, we should be aligning at a 1M or 2M boundary on the
root device, not within the partition. The offsets are supposed
to do that, and if there's a problem we should fix it.


> AHhhhh thought hits me... did the code get changed
> to make the images larger, and somehow the image
> creation code went from a 32 sector/track fake C/H/S
> to a 63 sector fake C/H/S, and the code has all
> along been assuming 32 sector/track drives?
>

63 sector for 'fake' C/H/S has been a thing since at least
FreeBSD 6 and maybe a little longer. There's no cutover
based on image size of the device. The older ATA code,
pre-cam but post SOS rewrite, would try very hard to return
the values from the underlying device that it reported. And that
would lead to mismatches because it was different than the lies
that md told by default. But that only mattered for really old
BIOSes that couldn't handle LBA/packet mode in booting (which
is the primary reason the old fdisk program could set the ending CHS
of partitions since the FreeBSD code used that to guess the CHS
of the drive itself in the absence of other information).

I don't think the kernel has changed in this area in a very long time.
At worse, we're seeing a mkimage bug.

Warner


> >
> > MBR vs. GPT is not the fundamental issue for that.
> >
> > ===
> > Mark Millard
> > marklmi at yahoo.com
> >
> >
> >
> >
>
> --
> Rod Grimes
> rgrimes@freebsd.org
>
>