[Cbe-oss-dev] Mini Req to PS3 Linux boot loader (Re: PS3 Linux boot loader discussions)

[Luca Barbato]

Hiroyuki Machida wrote:
> Fuse-san,
> 
> Thank you for correction and comment.
> 
> 2007/2/9, Hiroaki Fuse <Hiroaki_Fuse at hq.scei.sony.co.jp>:
>> I have some comments.
>>
>>  > = Minimum requiremnts to boot loader for supporting removable boot
>>  > media =
>>  >
>>  > == Background ==
>>  >
>>  > This document describes minimum requirements to boot loader for
>>  > PS3 Linux envrionment, so that system is able to boot from removable
>>  > media/devices.
>>  >
>>  > This document does NOT intend to describe requirements, so that
>>  > system is able to boot from internal HDD. It's up to the boot
>>  > loader's implementation, if it requires /etc/kboot.conf and/or
>>  > how the boot loader interprets it.
>>  > Because the contents of HDD could be changed later, to follow
>>  > newer bootloader configuration file.
>>  >
>>  > On the other hand, boot configuration file on removable boot
>>  > media like Live CD could be changed, once released.
>>                               ------
>>                               could not be ??
> 
> You're right.
> 
>>  > So we need minimum specification of the boot loader, so that
>>  > any released removable boot media stay to be bootable.
>>
>>
>>  >
>>  > === Minimum /etc/kboot.conf ===
>>  >
>>  > Bootloader needs to understand following minimum kboot.conf format.
>>  >
>>  > timeout=[num]
>>  > initrd=[path1]
>>  > root=[rootname]
>>  > default=[label]
>>  > [label]=[path2]  [kernel-params]
>>  >
>>  > - [num] is digit number; which describes value of boot
>>  > selection time out. It's just used as hint. Boot loader
>>  > ignore this value.
>>  > - [label] is a label for boot entry.
>>  > There's no alternative entry, we can omit "label" line like;
>>  > default=[path2] [kernel-params]
>>
>> I think that default must be a label.
>> When we insert multiple boot media, we have to select one of them.
> 
> 
> From the point of view of  the minimum specification,
> [label] looks not to be mandatory. On the other hand,
> I agree that [label] would help implmentation.
> 
> As you mention, in case of [label], it's expected to be uniq.

there weren't discussions about using some kind of hw signature
(depending on which hw it is) to avoid clashes around identical labels?

> Do you have any idea on max length of [label]  and limitation
> on  chars in [label], like "must be starting with alpahbet".
> 

IMHO A label should be at most fit in a single console line, so 79 chars
or even less if we assume we'd have to append the hw id in case of
clash. IF we have enough memory to support utf8 and think about fonts
availability, probably it could be nice, otherwise pure ascii could be a
simple constraint that avoids implementation headache.

lu

-- 

Luca Barbato

Gentoo/linux Gentoo/PPC
http://dev.gentoo.org/~lu_zero