> Why "just"? What else are you putting on the rom? My .trx is currently
> just under 2MB.
a 1.8MB .trx is already too big to be able to get a jffs partition.
My .trx is currently 1.9MB, including all the modules needed for a USB
root and an IDE root (tho I only care about the IDE root ones).
> Again, flexibility. What happens if your USB / is unavailable? Does
> your image boot successfully (i.e. failing through the mount_usb) or
> does the rom/jffs tradeoff cause an outright boot failure?
Well, strictly speaking, my hack is slightly inflexible, but only
slightly: it tries to mount a USB drive, then looks for /sbin/init and
if that works, use it. If it fails, it tries to mount the IDE drive
(hardcoded to /dev/hde since it's oddly the name used on my wl700ge),
and if that fails it keeps going with the default behavior, so it still
works just fine without a USB or IDE drive.
Of course, with 2MB of flash, the default behavior is pretty limited,
and with my large /rom, it mean the default is "rom+ram" rather than
"rom+jffs", so it's even more limited.
I think the choice about the order in which USB, IDE, and jffs are
attempted doesn't mater much. Except of course that my "try to mount
USB" includes a 10s wait. So my machine (which basically never has
a USB drive) always sits there for 10s during boot. It hasn't bother me
enough to change it yet.
>> IIRC even without those kernel modules, the jffs partition is either too
>> small to work
> Mine works just fine.
>> or it's too small to be convenient
> Mine is perfectly convenient.
Your machine has more than 2MB of flash, doesn't it?
>> (e.g. it kind of works but you get odd behaviors such as "ls
>> seg faults" and you need to use a crippled environment (busybox)).
> Not gotten anything like that at all with my images here.
Again: your machine has more than 2MB of flash, doesn't it?
If you look at the various "install kamikaze on your wl700ge" web pages,
you'll see several references to such "your system will be temporarilly
brittle, but trust us, it'll work in the end", with examples of ls
segfaulting (I've had other problems when trying their approach where
TAB completion would freeze the machine, from what I gather all those
problems are due to using jffs2 with too few eraseblocks, so it can
almost work right, but can't avoid tripping on itself every once in
a while).
>> You might be right. But I'm not sure what kind of flexibility you're
>> looking for here.
> Just the flexibility to put the same image on routers with or without
> USB storage hanging off of them.
I think the 2MB of flash of the wl-700ge makes it pretty much
unavoidable to use an ad-hoc image for it. For your machines, you can
probably use a common .trx without any significant downside.
> Yeah, I agree that's a huge problem. On the other hand, the USB / is
> not really upgradable anyway, given the limitations of ipkg.
Why not? I've upgraded lots of "core" packages already without too
much trouble. I do get into trouble with kernel modules, since ipkg
doesn't keep track of the kernel version of the module, so it thinks
I have kmod-usb2 installed as soon as it's installed for "some" kernel.
I'm sure there are snags (upgrading the base-files package is pretty
nasty, for example), but by and large, it's usable.
> The developers will tell you over and over again, the only way to
> really upgrade is to reflash. Apparently ipkg is not really robust
> enough to handle real upgrade scenarios. There are bugs in the Trac
> about ipkg refusing to believe that for a given package that version
> n+m (for some values of m and n) is an upgrade for n.
Yes, it's far from perfect.
> That said, I have toyed briefly with the thought of making openwrt work
> optionally and alternatively with dpkg. :-) I.e. you choose at build
> time whether you want to use ipkg or dpkg and when the image builds it
> works accordingly. Somehow Makefiles would have to create /debian dirs
> on the fly. I've not really thought it all through.
Another option is to install Debian on your USB /
I have a Debian chroot, but I've never tried to use it for /, really.
> FWIW, do you think my proposal to do the USB rootfs mount (assuming a
> USB storage device is present of course) *after* the jffs
> mount-and-pivot will actually break in your configuration?
No, it sounds fine.
Stefan
_______________________________________________
openwrt-devel mailing list
openwrt-devel@lists.openwrt.org
http://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-devel
