binman warning/failure when blobs are missing

[Tom Rini]

On Thu, Aug 11, 2022 at 06:08:51PM -0600, Simon Glass wrote:
> Hi,
> 
> On Thu, 11 Aug 2022 at 11:44, Tom Rini <trini at konsulko.com> wrote:
> >
> > On Thu, Aug 11, 2022 at 09:59:05AM -0700, Tim Harvey wrote:
> > > On Thu, Aug 11, 2022 at 9:48 AM Tom Rini <trini at konsulko.com> wrote:
> > > >
> > > > On Thu, Aug 11, 2022 at 09:27:44AM -0700, Tim Harvey wrote:
> > > >
> > > > > Greetings,
> > > > >
> > > > > After a couple of hours troubleshooting a bad boot image today I
> > > > > realized the issue was that I had some 0 byte files for the lpddr4
> > > > > training blobs that are part of the imx8mp binman created image.
> > > > >
> > > > > Digging in I found that if a blob referenced in the binman node is
> > > > > missing a warning will be output but the missing files will be
> > > > > 'created' as 0 byte files such that the next time you build you will
> > > > > get no warning (but will have a non-working image). Additionally the
> > > > > error does not cause a non-zero exit code.
> > > > >
> > > > > I'm not that fluent in python these days, and don't have the time for
> > > > > a while to try and fix this but I figured I would at least send this
> > > > > email in case someone else does.
> > > > >
> > > > > Example:
> > > > > # rm *lpddr4*.bin # make sure lpddr4*.bin files referenced in binman
> > > > > nodes are missing
> > > > > # make distclean imx8mp_venice_defconfig flash.bin && echo "build ok"
> > > > > ...
> > > > >   BINMAN  flash.bin
> > > > > Image 'main-section' is missing external blobs and is non-functional: ddr-1d-ime
> > > > > m-fw ddr-1d-dmem-fw ddr-2d-imem-fw ddr-2d-dmem-fw
> > > > > Image 'main-section' has faked external blobs and is non-functional: lpddr4_pmu_
> > > > > train_1d_imem_202006.bin lpddr4_pmu_train_1d_dmem_202006.bin lpddr4_pmu_train_2d
> > > > > _imem_202006.bin lpddr4_pmu_train_2d_dmem_202006.bin
> > > > >
> > > > > Some images are invalid
> > > > > ^^^ excellent warning
> > > > > build ok
> > > > > ^^^ not so great that there is a successful exit code
> > > > > # make flash.bin && echo "build ok"
> > > > > ...
> > > > >   BINMAN  flash.bin
> > > > > build ok
> > > > > ^^^ absolutely horrible that 0 byte files were created and thus
> > > > > everything looks good this time around!
> > > > > # stat -c "%s %n" lpddr4*.bin
> > > > > 0 lpddr4_pmu_train_1d_dmem_202006.bin
> > > > > 0 lpddr4_pmu_train_1d_imem_202006.bin
> > > > > 0 lpddr4_pmu_train_2d_dmem_202006.bin
> > > > > 0 lpddr4_pmu_train_2d_imem_202006.bin
> > > >
> > > > So, this isn't the first time someone has had this problem. On the one
> > > > hand, we need CI to pass, and not require fetching of arbitrary further
> > > > images to assemble the binary. On the other hand, we don't want users
> > > > spending a bunch of time because something didn't work and the normal
> > > > way of conveying THIS WON'T WORK is a non-zero exit status. Can we
> > > > easily make some flag for buildman or binman that we do set in CI but
> > > > won't be set by users?
> > > >
> > >
> > > Tom,
> > >
> > > Ok - that makes sense as far as the exit status goes. It would be MUCH
> > > easier to catch an error like this if binman didn't create 0 byte
> > > files for missing files so that you at least get the (ascii colored)
> > > message indicating the image is wrong.
> 
> Please see this:
> 
> https://patchwork.ozlabs.org/project/uboot/patch/20220807154708.1418967-2-sjg@chromium.org/
> 
> > >
> > > I do love the idea of a flag for CI also!
> >
> > So, that's part of the root of the problem here.  We're passing
> > --fake-ext-blobs to binman so that it will create those 0 byte files
> > and it in turn complain. I think we didn't figure out a good way to
> > tell buildman to pass that to binman tho?
> 
> I hope the above patch will fix this part of the problem.

I'm not sure. What I really want to see is something like buildman
--fake-ext-blobs, which in turn does 'make BINMANFLAGS=--fake-ext-blobs'
or 'make BINMAN_FAKE_EXT_BLOBS=1' and that is what sets --fake-ext-blobs
to be passed inside of cmd_binman, so that we can tell CI to do that
(just like we do -E today, to ensure -Werror) but regular users get
build failures.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20220811/c33132e6/attachment.sig>