Re: What is this grub_disk_read doing in the i386 linux loader?

[Daniel Kiper]

On Tue, Apr 24, 2018 at 04:08:57PM -0700, Andrew Jeddeloh wrote:
> Thanks for the reply.
>
> I'm not sure I follow. Looking over the 32 bit boot spec, it looks
> like the process is:
>
> 1) zero out linux_params
>  - grub does this
> 2) copy the linux boot params (from 0x1f1) into linux params
>  - grub does this by reading from 0x0 until the end of lh, then
> copying lh+0x1f1 til the end of lh into linux_headers [4][5]
> 3) Do the read/write/modify operations that would happen for a 16 bit boot
> 4) calculate the end of the the setup header
>  - grub does not do this and I think assumes its just the end of the
> linux_params struct

As Michael said, "0x0202 + byte value at offset 0x0201" is your friend.

> 5) fill out the rest of the setup header things from the zero page doc [6]
>  - AFAICT this isn't meant to be read from the kernel image, but
> instead filled out by bootloader or left as zero

AIUI it is. Please look above.

>    - It looks like the current code is trying to do this but is just
> assuming the end is at the end of linux_params.

This is wrong due to grub_e820_mmap.

>  - Looking over the items in the zero page doc, there doesn't seem to
> be anything that the kernel could supply a useful "default" value.

GRUB have to load the rest of Linux header into linux_params.
Current math is wrong and have to be fixed. Please look above
and think about newer versions of Linux headers which GRUB is
not aware of.

> As I understand it, the read is not harmful (as it just gets
> overwritten by the bootloader later) but also not needed. Is this your
> understanding?

This read is needed, however, math has to be fixed. Additionally,
I think that you can take a look at SYSLINUX and do something
similar in GRUB.

> I'll definitely submit a patch upstream once I figure out what should
> be going on.

Great! Thanks!

Daniel