grub-install

grub-install

[Yoshinori K. Okuji]

I have mostly implemented grub-install as a shell script. I just copied it 
from GRUB Legacy then adapted it to GRUB 2. This script is much simpler in 
GRUB 2, because grub-setup handles most things.

It has one remaining issue, and I'm not sure what is appropriate for this.  
Since the core image in GRUB 2 is generated dynamically, it is necessary to 
specify what modules should be included in it. When the user uses 
grub-install, I believe that they should be automatically determined.

For now, pc.mod and _chain.mod are hardcoded. I think these are good enough 
for PC. It might be better to add _linux.mod on GNU/Linux, though. I'm not 
sure.

The issue is how to determine a filesystem module which is needed to load more 
modules. I think there are three ways:

1. Shell scripting. Probably it is possible to write a shell function which 
detects a filesystem by using df, mount, etc. I do not like this approach 
very much, because the installation environment can be very minimalistic 
(suppose when you install an OS using a floppy), and the behaviors of 
commands are sometimes not clever enough.

2. Adding an option into grub-mkimage to add a filesystem module 
automatically, depending on the filesystem of a grub directory. This is not  
bad, but this makes grub-mkimage a bit complex.

3. Adding a new utility which detects an appropriate filesystem module for a 
given directory. This is not bad, but this sounds a bit overkill.

What do you think?

Okuji

Re: grub-install

[Hollis Blanchard]

On Jul 24, 2005, at 2:44 PM, Yoshinori K. Okuji wrote:
> I have mostly implemented grub-install as a shell script. I just 
> copied it
> from GRUB Legacy then adapted it to GRUB 2. This script is much 
> simpler in
> GRUB 2, because grub-setup handles most things.

Great! This is something we still need to do on PowerPC before grub2 is 
generally usable...

> It has one remaining issue, and I'm not sure what is appropriate for 
> this.
> Since the core image in GRUB 2 is generated dynamically, it is 
> necessary to
> specify what modules should be included in it. When the user uses
> grub-install, I believe that they should be automatically determined.

Is there a reason not to add all modules all the time?

-Hollis

Re: grub-install

[Yoshinori K. Okuji]

On Monday 25 July 2005 04:04, Hollis Blanchard wrote:
> Is there a reason not to add all modules all the time?

I should have written more info in English. This is documented well, but only 
in Japanese.

There are several reasons. In the context of i386-pc:

- The size problem. We must keep the core image at least 31KB, so that we can 
embed it into the space right after the MBR. This is the most important 
reason. Very specific to i386-pc. I have no idea if other architectures have 
similar restrictions, but probably they do.

- The loading time. Potentially, the number of modules is infinite in GRUB 2.

- Conflicts. For now, we do not have multiple modules which provide the same 
functionality, but it is one of the goals of modules to make this possible.

- Runtime slowness. For example, if I load amiga partition map support, this 
makes apparently it slower to access partitions in QEMU.

- Hardware probing. In particular, probing ISA ethernet devices which are not 
present in a system frequently causes a system crash or a too long timeout.

Actually, many reasons are the same as "why we don't make initrd with all 
modules of linux" or "why we don't build all features into linux".

Okuji

Re: grub-install

[P]

[-- Attachment #1: Type: text/plain, Size: 1114 bytes --]

Yoshinori K. Okuji wrote:
> On Monday 25 July 2005 04:04, Hollis Blanchard wrote:
> 
>>Is there a reason not to add all modules all the time?
> 
> 
> I should have written more info in English. This is documented well, but only 
> in Japanese.
> 
> There are several reasons. In the context of i386-pc:
> 
> - The size problem. We must keep the core image at least 31KB, so that we can 
> embed it into the space right after the MBR. This is the most important 
> reason. Very specific to i386-pc. I have no idea if other architectures have 
> similar restrictions, but probably they do.

In case others don't know the reason for "31KB":

As can be seen in the attached common linux disk layout diagram,
the optional DOS compatability region used by stage 1.5 is added by
default by most partition managers, and is there so that partitions
start on a cylinder boundry. Therefore its size is determined by the
number of sectors (512 bytes) per cylinder. The maximum sectors per
cylinder is 64, and so the largest size available for grub's stage 1.5
is 63 * 512, or 32256 bytes.

Pádraig.

[-- Attachment #2: grub.png --]
[-- Type: image/png, Size: 23224 bytes --]

[-- Attachment #3: grub.dia --]
[-- Type: application/x-dia-diagram, Size: 2141 bytes --]

Re: grub-install

[Marco Gerards]

"Yoshinori K. Okuji" <okuji@enbug.org> writes:

> 3. Adding a new utility which detects an appropriate filesystem module for a 
> given directory. This is not bad, but this sounds a bit overkill.

It's not *that* hard to write.  Just probe the filesystem using
modules until one works.  For that module loading is required which
can be done in GNU/Linux.  The main problem is that you need to use
mmap to get executable memory (which is OS specific) and the other
problem are the relocations, the allocated memory needs to be close to
the executable on the PPC.

How is the correct stage1.5 selected in GRUB Legacy?

--
Marco

Re: grub-install

[Vincent Pelletier]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=us-ascii, Size: 1285 bytes --]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Yoshinori K. Okuji wrote:
> When the user uses 
> grub-install, I believe that they should be automatically determined.

...while the choice could be completely overridden, and maybe making
possible to choose manualy some modules.

> For now, pc.mod and _chain.mod are hardcoded. I think these are good enough 
> for PC. It might be better to add _linux.mod on GNU/Linux, though. I'm not 
> sure.

Good idea to add _linux.mod, if it (adding) can be disabled.

> 3. Adding a new utility which detects an appropriate filesystem module for a 
> given directory. This is not bad, but this sounds a bit overkill.

I prefer that solution. I think grub-mkimage should only 'know' how to
make an image from modules, another tool telling it which ones to add.
And I think that "other tool" should be grub-install.

Vincent Pelletier
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFC5OPwFEQoKRQyjtURAo+1AJ9OLTTDRAvJRJ2rNiu0rg3sXkmOMwCZAXhT
+eVXyfuvUtsGsxSZJSHdeg0=
=GNoV
-----END PGP SIGNATURE-----

	

	
		
___________________________________________________________________________ 
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger 
Téléchargez cette version sur http://fr.messenger.yahoo.com

Re: grub-install

[P]

P@draigBrady.com wrote:
> As can be seen in the attached common linux disk layout diagram,

sigh, that 16KiB in the diagram should be 32KiB.
Anyway you get the idea...

-- 
Pádraig Brady - http://www.pixelbeat.org
--

Re: grub-install

[Hollis Blanchard]

On Jul 25, 2005, at 6:26 AM, P@draigBrady.com wrote:

> Yoshinori K. Okuji wrote:
>> On Monday 25 July 2005 04:04, Hollis Blanchard wrote:
>>> Is there a reason not to add all modules all the time?
>> I should have written more info in English. This is documented well, 
>> but only in Japanese.
>> There are several reasons. In the context of i386-pc:
>> - The size problem. We must keep the core image at least 31KB, so 
>> that we can embed it into the space right after the MBR. This is the 
>> most important reason. Very specific to i386-pc. I have no idea if 
>> other architectures have similar restrictions, but probably they do.
>
> In case others don't know the reason for "31KB":
>
> As can be seen in the attached common linux disk layout diagram,
> the optional DOS compatability region used by stage 1.5 is added by
> default by most partition managers, and is there so that partitions
> start on a cylinder boundry. Therefore its size is determined by the
> number of sectors (512 bytes) per cylinder. The maximum sectors per
> cylinder is 64, and so the largest size available for grub's stage 1.5
> is 63 * 512, or 32256 bytes.

This is for GRUB Legacy, right? There is no "stage 1.5" In GRUB2? What 
is the new layout?

-Hollis

Re: grub-install

[Marco Gerards]

Hollis Blanchard <hollis@penguinppc.org> writes:

>> In case others don't know the reason for "31KB":
>>
>> As can be seen in the attached common linux disk layout diagram,
>> the optional DOS compatability region used by stage 1.5 is added by
>> default by most partition managers, and is there so that partitions
>> start on a cylinder boundry. Therefore its size is determined by the
>> number of sectors (512 bytes) per cylinder. The maximum sectors per
>> cylinder is 64, and so the largest size available for grub's stage 1.5
>> is 63 * 512, or 32256 bytes.
>
> This is for GRUB Legacy, right? There is no "stage 1.5" In GRUB2? What
> is the new layout?

You are right.  In GRUB 2 there is no stage 1.5 and stage2.  In GRUB 2
there is a core image.  It mainly consists of the stuff in kern/, some
stuff from disk/i386/pc, term/i386/pc.  Using that is can load
modules.

The core image with some modules replace stage 1.5.  So the main core
of GRUB together with some important modules are stored in this 31KB.
Because this space is limited not every module can fit in there.  Most
importantly is the filesystem module.  It is used to load other
modules from that filesystem.

So an example of a boot process is:

- The BIOS loads the GRUB2 (boot.img) as it is stored in the MBR

- GRUB2 in the MBR loads that 31KB into memory and jumps to it

- That 31KB consists of the kernel (kernel.img) and some raw modules

- The kernel (rescue mode) starts and loads every module

- Rescue mode tries to load normal.mod

- If normal.mod can be found on the prefix path it is loaded

- Normal mode looks for grub.cfg on the prefix path and loads it

- The menu is shown and the user can choose an OS to boot

It is possible to load additional modules so other filesystems can be
accessed, other loaders can be used, etc.  On the PC a nice set of
modules is: pc.mod (for partition access), _chain.mod (rescue mode
chainloader), ext2.mod.  Of course ext2.mod is for the case that you
want to load normal.mod and other modules from an ext2 filesystem.

I hope this e-mail made clear how GRUB 2 basically works on the PC.
I am not a PC expert and my English is not that good.  Hopefully
someone can and will correct me if I was wrong so I will understand
things better.  Hopefully Hollis' question is answered.

Thanks,
Marco

Re: grub-install

[Yoshinori K. Okuji]

On Monday 25 July 2005 15:06, Vincent Pelletier wrote:
> ...while the choice could be completely overridden, and maybe making
> possible to choose manualy some modules.

Good idea.

> Good idea to add _linux.mod, if it (adding) can be disabled.

OK.

> I prefer that solution. I think grub-mkimage should only 'know' how to
> make an image from modules, another tool telling it which ones to add.
> And I think that "other tool" should be grub-install.

OK. That sounds reasonable. I follow this.

Okuji

Re: grub-install

[Yoshinori K. Okuji]

On Monday 25 July 2005 18:35, Marco Gerards wrote:
> So an example of a boot process is:
>
> - The BIOS loads the GRUB2 (boot.img) as it is stored in the MBR
>
> - GRUB2 in the MBR loads that 31KB into memory and jumps to it
>
> - That 31KB consists of the kernel (kernel.img) and some raw modules

In details, the core image consists of diskboot.img, kernel.img then zero or 
more modules. boot.img loads only the first sector (diskboot.img) from a 
disk, and diskboot.img loads the rest of the core image.

In the case of PXE, pxeboot.img is prepended to the core image, and simply 
jumps to the start address in kernel.img, because PXE ROM loads the whole 
core image from a network.

> - The kernel (rescue mode) starts and loads every module

The kernel initializes the system briefly, and decompresses the compressed 
part of the core image. After this, the kernel loads all pre-loaded modules 
from the memory.

> I hope this e-mail made clear how GRUB 2 basically works on the PC.
> I am not a PC expert and my English is not that good.  Hopefully
> someone can and will correct me if I was wrong so I will understand
> things better.  Hopefully Hollis' question is answered.

Your description is really good.

Well, the bootstrap on i386-pc is complicated, because PC BIOS sucks. On other 
architectures, this can be simplified very much.

Okuji

Re: grub-install

[Yoshinori K. Okuji]

On Monday 25 July 2005 14:15, Marco Gerards wrote:
> It's not *that* hard to write.

Not hard at all. It is a design problem but not an implementation problem.

> How is the correct stage1.5 selected in GRUB Legacy?

Very ugly. The function install_func includes a list of pairs, where each pair 
consists of a filesystem name and a Stage 1.5 name.

Okuji

Re: grub-install

[P]

P@draigBrady.com wrote:
> P@draigBrady.com wrote:
> 
>> As can be seen in the attached common linux disk layout diagram,
> 
> 
> sigh, that 16KiB in the diagram should be 32KiB.
> Anyway you get the idea...

Actually do we have 32,256 bytes to use?
I was wondering about sectors 62 & 63 (counting from 0).
On my laptop (that had windows wiped the minute I got it),
I notice that there is a MSWIN4.1 boot record in sector 63
(I think it actually flows over the next 2 sectors also).
So can we generally use that? I.E. does grub provide all
the functionality of that sector?
Also sector 62 has the following? Is that BIOS suspend
to disk info or something, or maybe something left
over from the windos install?

# dd count=1 skip=62 if=/dev/hda 2>/dev/null | od -Ax -tx1z
000000 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  >................<
*
0000c0 67 68 46 50 b6 96 ca 3e bb 35 eb 3d 32 54 4f 53  >ghFP...>.5.=2TOS<
0000d0 48 49 42 41 00 33 00 00 00 00 00 00 00 00 00 00  >HIBA.3..........<
0000e0 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  >................<
*
000100 00 00 00 00 00 00 00 00 00 34 33 4d 35 33 36 38  >.........43M5368<
000110 39 54 00 00 00 00 00 00 00 00 00 00 00 54 4f 53  >9T...........TOS<
000120 48 49 42 41 20 4d 4b 32 30 31 38 47 41 53 00 00  >HIBA MK2018GAS..<
000130 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  >................<
*
000200


-- 
Pádraig Brady - http://www.pixelbeat.org
--

Re: grub-install

[Vincent Pelletier]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Yoshinori K. Okuji wrote:
> Well, the bootstrap on i386-pc is complicated, because PC BIOS sucks. On other 
> architectures, this can be simplified very much.

I'm afraid a stage 1 -like thing will be needed on usparc. I have found
few docs about the actual implementation of the disk boot process on
usparc, but it sounds like "load the second sector of disk, and run it"
(the first sectors is filled with the partition table). I still have no
success in making a bot floppy "by hand" (parted/fdisk + dd). I get a
magic number error, while the crc in the partition table should be
correct. I'll check that when I have more time, in 3~4 weeks.

Vincent Pelletier
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFC5oPlFEQoKRQyjtURAmpjAJ98vU2sF3Gb+P/sLbrxu8lLCkktuACgkGKr
H5u+iuo/4eX1Ui91mJJxjXw=
=6LrZ
-----END PGP SIGNATURE-----

	

	
		
___________________________________________________________________________ 
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger 
Téléchargez cette version sur http://fr.messenger.yahoo.com

Re: grub-install

[Yoshinori K. Okuji]

On Tuesday 26 July 2005 15:13, P@draigBrady.com wrote:
> I was wondering about sectors 62 & 63 (counting from 0).
> On my laptop (that had windows wiped the minute I got it),
> I notice that there is a MSWIN4.1 boot record in sector 63
> (I think it actually flows over the next 2 sectors also).

It is just a trace of the Windows boot sector. Once you remove Windows, it is 
a garbage.

> Also sector 62 has the following? Is that BIOS suspend
> to disk info or something, or maybe something left
> over from the windos install?

I don't know. If you are afraid, you can try to fill that with zero (after 
making a backup) and see what happens.

Okuji

Re: grub-install

[Yoshinori K. Okuji]

On Tuesday 26 July 2005 20:41, Vincent Pelletier wrote:
> I'm afraid a stage 1 -like thing will be needed on usparc.

You are right. On SparcStation, the firmware loads only 512 bytes like PC as a 
boot sector. So you need to make boot.img separately. The difference is that 
there is no "safe place" in SparcStation, and you can choose a sector to be 
loaded in the firmware. You can get some information from the web site of 
SILO:

http://www.sparc-boot.org/how.html

> I have found 
> few docs about the actual implementation of the disk boot process on
> usparc, but it sounds like "load the second sector of disk, and run it"
> (the first sectors is filled with the partition table). I still have no
> success in making a bot floppy "by hand" (parted/fdisk + dd). I get a
> magic number error, while the crc in the partition table should be
> correct. I'll check that when I have more time, in 3~4 weeks.

Which model do you have? I hear that most UltraSparc models cannot boot from a 
floppy due to a bug in OpenBoot. Therefore, when I installed GNU/Linux into a 
UltraSparc machine, I used netboot.

Okuji

Re: grub-install

[Vincent Pelletier]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Yoshinori K. Okuji wrote:
> Which model do you have? I hear that most UltraSparc models cannot boot from a 
> floppy due to a bug in OpenBoot. Therefore, when I installed GNU/Linux into a 
> UltraSparc machine, I used netboot.

I have an Ultra 10 (looking for an U80, with 4*450Mhz USparc II... quite
expensive, but I can't get it out of my head :) ). I've heard of this
bug, I'll have to test it with a reference floppy. I did some tests with
a spare HD I used to hack grub 2, and I had the same error, but it might
be another problem.
I used netboot to start the port, it works great with ELF64 unstripped
binaries.

Vincent Pelletier
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQFC5pWCFEQoKRQyjtURAmV6AJ9ncgVwA8ezoZ1BD3snBzl06+9Z5ACgouLF
IvakTTvdLqFUGLfkWTZ5zUU=
=tFDa
-----END PGP SIGNATURE-----

	

	
		
___________________________________________________________________________ 
Appel audio GRATUIT partout dans le monde avec le nouveau Yahoo! Messenger 
Téléchargez cette version sur http://fr.messenger.yahoo.com

grub-install

[E.R]

[-- Attachment #1: Type: text/plain, Size: 108 bytes --]

hi!
i can't install grub2 into sdb hard, it install in sda but dont install in
sdb, how to do  it?

-- 
e.r

[-- Attachment #2: Type: text/html, Size: 141 bytes --]

Re: grub-install

[Pavel Roskin]

On Tue, 2008-05-27 at 09:36 +0430, E.R wrote:
> hi!
> i can't install grub2 into sdb hard, it install in sda but dont
> install in sdb, how to do  it?

You need to mount one of the partitions on /dev/sdb and specify it in
the --root-directory:

mount /dev/sdb1 /mnt/tmp
grub-install --root-directory=/mnt/tmp /dev/sdb
umount /mnt/tmp

In the future, please provide more details, such as the exact command
line, the exact error messages and the version of GRUB.

-- 
Regards,
Pavel Roskin

Re: grub-install

[Gregg Levine]

On Tue, May 27, 2008 at 1:06 AM, E. R <elhamrastgoo@gmail.com> wrote:
> hi!
> i can't install grub2 into sdb hard, it install in sda but dont install in
> sdb, how to do  it?
>
> --
> e.r
> _______________________________________________
> Grub-devel mailing list
> Grub-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/grub-devel
>
>
(Reposting my earlier message from another known working ID because it
hasn't arrived from my regular one. Please complain directly if it
finally does regarding duplicates.)

Hello!
Instead of repeating the earlier question, why not present us with
more information on your setup?

Also please try not to use the Google Mail HTML mail settings for all
communications, it is not the default setting, plain text is. This
list, and all for the GNU lists, prefer to use plain text e-mail.
-- 
Gregg C Levine gregg.drwho8@gmail.com
"This signature was once found posting rude
 messages in English in the Moscow subway."

grub-install

[E.R]

[-- Attachment #1: Type: text/plain, Size: 556 bytes --]

hi!
very thnx Pavel Roskin, i do :

mount /dev/sdb1 /mnt/tmp
grub-install --root-directory=/mnt/tmp /dev/sdb
umount  /mnt/tmp

but grub2 don't install in sdb, i have 2 hard, sda & sdb , & my linux is
into sdb. i want to change my grub-0.97  to grub-1.96, i bulid grub-1.96.rpm
and install it with command :

rpm -ivh grub-1.96.rpm

, but it don't work in sdb . so i install grub-1.96.tar.gz in sdb and use :
grub-install /dev/sda , i see this error:

*welcome in grub!
Enterin into rescue mode ...
grub rescue >

*again thnx a lot for your answer

-- 
e.r

[-- Attachment #2: Type: text/html, Size: 672 bytes --]