On 8 December 2011 11:54, Dave Pigott <dave.pigott@linaro.org> wrote:

Just wanted to add a couple of comments:

Zygmunt commented earlier that creating master images is hard. I actually

disagree, but that is all about perspective and familiarity. I create master

images on a *very* regular basis and it's pretty easy once you know what

you're doing, but that's the point. It's about what you're most familiar

with.

Is the official documentation to create a master image available on
https://wiki.linaro.org/Platform/Validation/Specs/MasterBootImage ?
If it's the case, we should convert it to a step by step how to and
link it to Validation knowledge base.

+1 on Spring's idea that we should get l-m-c to create a tailored,

partitioned image. I've thought the same for a while, but since it's

application was only for LAVA I felt it would be a far lower priority than

many other things.

Also +1 on the longer term strategy of supporting NBD (or whatever) type

solutions.

Like others in this thread, I'm also concerned about the attempt to

reproduce conmux within LAVA. Yes, we get rid of a lot of nasty Perl code

(in case I offend, which I don't intend to, I mean Perl is nasty, not the

implementation,) but we have to consider whether the effort involved would

significantly benefit us in the end result.

Additionally, the idea of not supporting the serial interface, which has the

distinct advantage of capturing the log all in one place, I find difficult

to justify. Whilst I'm not completely wedded to it, as Ricardo said, it is

the one interface that is guaranteed to be provided across all boards in one

form or another. If we'd not had it, we would be significantly further

behind in deploying origens right now.

++2c

Dave Pigott

Validation Engineer

T: +44 1223 45 00 24 | M +44 7940 45 93 44

Linaro.org │ Open source software for ARM SoCs

Follow Linaro: Facebook | Twitter | Blog

On 8 Dec 2011, at 09:39, Spring Zhang wrote:

Generally, there are two ways to drive the boards: by serial line or

network(NFS maybe plus NBD). If all switching to NFS, I'm sure there will be

other problems in daily using.

For the MAC address issue, it will occur both on current master/tester image

layout and NFS deployment if uboot can't get board fused MAC address and

pass it to kernel.

For the master image generation, I think the best way is to make LMC be able

to create a custom master SD card layout, now it only needs to extend to

shrink boot and rootfs partition, and make two or more blank

partitions(testboot and testrootfs) with label in one command.

--

Best wishes,

Spring Zhang