Does anyone tftpboot the userspace image?

[Xo Wang]

Hi folks,

Thanks for chipping in on this.

On Sun, Jan 22, 2017 at 5:10 PM, Brad Bishop
<bradleyb at fuzziesquirrel.com> wrote:
> Hi Xo
>
> Honestly this is my _only_ use case… I don’t ever flash machines unless
> whatever I am testing specifically requires that, which for me anyway,
> is very infrequent.
>
> There doesn’t necessarily have to be an all or nothing approach taken here.
>
> cpio.lzma.u-boot can be removed from IMAGE_FSTYPES in several places:
>
> -in your local.conf
> -in zaius.conf in the zaius layer
> -in a yet to be created ingrasys.conf in the ingrasys layer
> -probably other ways too
>
> Alternatively, we can turn it off by default and add it where it is desired - possibly
> -in my local.conf
> -in in ibm.conf
>

I don't think we have to do anything about this. I didn't realize what
the images are useful for until after I had mailed the change.

Now that I know netbooting from the rootfs is possible I'll try it on
my setup. It should save a fair bit of time flashing the userspace
portion---another development annoyance for me. :)

You should add this to the cheatsheet in the project docs.

> I guess what I am not sure of is which makes more sense to be the default?
> When these kinds of questions come up I tend to think in terms of what would
> benefit new users of the project the most?  I honestly don’t know - what does
> everyone else think?
>

I would say that it's a fair default to generate all the useful
products on a simple "bitbake obmc-phosphor-image" invocation. Like
you said, it's not difficult to remove products in local.conf.

A third solution is to fix up the build dependency chain:
do_generate_flash currently depends on do_image_complete (thus all the
fs image products). If the netboot image were a separate recipe from
obmc-phosphor-image, then you could run "bitbake -c generate_flash
obmc-phosphor-image" for only the assembled flash image.

However, I'm pretty sure this would require two image roots be
installed with the same packages, so a normal build would be a lot
slower. Probably not a good solution, but maybe you folks know of a
way to make that work.

> -brad
>
>
>> On Jan 20, 2017, at 8:59 PM, Xo Wang <xow at google.com> wrote:
>>
>> Hi folks,
>>
>> I noticed from this discussion
>> https://lists.ozlabs.org/pipermail/openbmc/2016-April/thread.html#2738
>> that kernel developers were tftpbooting the userspace image from a
>> obmc-phosphor-image-<machine>.cpio.lzma.u-boot file.
>>
>> 1. How does/did that work? I guess you needed a custom init in the
>> initrd to load the u-boot container (?) instead of from mtd?
>>
>> 2. Are you still using this? Building the extra .cpio.lzma.u-boot is
>> kind of slow, with an extra ~45 seconds to do 'find | cpio | lzma;
>> mkimage' every build, and it can't be parallelized.
>>
>> I ask because I mailed a change to remove those image products from
>> the build, then followed the 'blame' to this use case:
>> https://gerrit.openbmc-project.xyz/#/c/1957/
>>
>> cheers
>> xo
>> _______________________________________________
>> openbmc mailing list
>> openbmc at lists.ozlabs.org
>> https://lists.ozlabs.org/listinfo/openbmc

cheers
xo