6 MB limit of initrd

Doug Maxey dwm at austin.ibm.com
Thu Feb 7 05:14:13 EST 2008


On Wed, 06 Feb 2008 16:07:46 GMT, Paul Nasrat wrote:
>
>On 6 Feb 2008, at 15:54, Gerhard Tonn wrote:
>
>> Scott Moser schrieb:
>>> On Fri, 1 Feb 2008, Gerhard Tonn wrote:
>>>
>>>
>>>> Hi,
>>>>
>>>> I am facing the problem that yaboot can't handle an initrd of more  
>>>> than
>>>> 6 MB. I tried the yaboot from head from
>>>> http://people.fedoraproject.org/~pnasrat/yaboot , since I read on
>>>> -devel, that the 6 MB problem has been fixed in head.
>>>> <http://people.fedoraproject.org/%7Epnasrat/yaboot>
>>>>
>>>> When running this yaboot, I get
>>>>
>>>> Claim failed for initrd memory
>>>> ramdisk load failed !
>>>> ENTER called ok
>>>>
>>>> Anybody any idea?
>>>>
>>>
>>> Please try building with commit  
>>> c8b04c614770870d75825c19bcd4729cc779b84a
>>> backed out (ie: git-checkout -f; git-show c8b04c61 | patch -p1 -R ;  
>>> make
>>> clean ; make
>>>
>>> I have a colleague who was having a similar problem on a power5 lpar
>>> system with firmware at SF240_332 .  Doing the above fixed the  
>>> problem.
>>> He's now able to netboot 6MB+ kernels.
>>>
>>>
>>>
>> Thanks, that helps indeed.
>
>Thanks for the confirmation and the patch, I'll revert and apply your  
>patch tonight.
>
>I'll try chase up Leo to find out what systems were tested with that  
>commit.

Shooting from hip here, and talking about *only* POWER5 and later machines.
Just from experience, I think the biggest variable is the version of the
firmware vs the machine type.  :) 

Historically, most IBM FW builds were machine specific forks from a previous 
machine type.   

In the recent past, for POWER5 and later machines, a change was made to
a more library oriented build.  Therefore changes in common areas of the
code would get automatically picked up for other systems during the next
build cycle.  One of the most visible changes was a common network stack,
which extends the parameters while on certain systems, enables iscsi boot.

I don't have any version specific info that might affect the max load
size, but am sure that FW from the last 6 months or so would examine
the size and or load-address of the kernel, and if different from the
previous version of the kernel, it would stop and reboot the system to
move the FW load address to accomodate.

It would sorta look like a crash and a silent reboot, but on the second
pass the boot would proceed normally.  If you weren't watching, you
might just think the boot was just taking extra long.

++doug





More information about the Yaboot-users mailing list