Fwd: Hello -- NAND/NOR Flash design for GTA02

Rod Whitby rod at whitby.id.au
Fri Jan 11 21:53:00 CET 2008

Have you guys considered a combination of a read-only rootfs (in squashfs or romfs) and a read-write jffs2 overlay using mini_fo, unionfs or some other overlay method?
OpenWrt does this, and they do *all* the booting from the read-only part with a tmpfs overlay, and then swap the tmpfs and jffs2 overlay after the jffs2 has been mounted in the background.
Fast boot, single jffs2, no change to packaging.
-- Rod (top-posting from Treo650 - sorry to those it offends)

-----Original Message-----
From: Andy Green <andy at openmoko.com>
Date: Friday, Jan 11, 2008 10:23 pm
Subject: Re: Fwd: Hello -- NAND/NOR Flash design for GTA02
To: Miles Hsieh <miles_hsieh at openmoko.com>
CC: Willie <willie_chen at openmoko.com>, openmoko-kernel at lists.openmoko.org,	Jim Huang <jserv at openmoko.org>,	=?UTF-8?B?IidNSUxFU19IU0lFSCAo6Kyd5Luy5YexKSci?=	<MILES_HSIEH at fic.com.tw>, nod <nod_huang at openmoko.com>,	"Neng-Yu Tu \(Tony Tu\)" <tony_tu at fic.com.tw>

>Hash: SHA1
>Somebody in the thread at some point said:
> Hi, Werner, Willie, Nod:
> production information now:
> device nand0 <neo1973-nand>, # parts = 6
> #: name                        size            offset          mask_flags
> 0: u-boot              0x00040000      0x00000000      0
> 1: u-boot_env          0x00040000      0x00040000      0
> 2: kernel              0x00800000      0x00080000      0
> 3: splash              0x000a0000      0x00880000      0
> 4: factory             0x00040000      0x00920000      0
> 5: rootfs              0x076a0000      0x00960000      0
>Okay by me, but one thing... there is a 20s target now for boot time. We have to decide what to do to get ourselves much closer to this (I
>don't think we will reach it) because we are far away from it right now.
>One of the choices that we may do is to cut rootfs into two, because the amount of time spent mounting is proportional only to the partition
>size.  We can create a 50MB "early boot" partition we can mount to get started with the initscripts, and mount the remaining 200MB in the
>background at a different mountpoint.  This will reduce the effective initial JFFS2 mount time to 20% of what it is now.
>Werner is looking at trying to find a different way to optimize the mount time, we said that early next week we will finalize the decision on the partition set we will use, but I would propose we cut the rootfs in two as described if Werner didn't have luck in speeding it up by a considerable amount.  Of course even if he speeds it up by 5% that is really good either way.
>If we go that way, we need to consider the impact on packaging paths. Probably we should leave all the packages to point where they are, and have /usr and so on in the "early boot" partition.  Then other packages should place stuff in something like /opt or /usr/local which is mounted or mounted --bind to the second, larger partition.
>- -Andy
>Version: GnuPG v1.4.7 (GNU/Linux)
>Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

More information about the openmoko-kernel mailing list