Re: [PATCH] Revert "tree: imx: remove old fit generator script"

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 10205 bytes --]

On Fri, Jan 07, 2022 at 12:21:18PM -0600, Adam Ford wrote:
> On Fri, Jan 7, 2022 at 11:38 AM Tom Rini <trini@konsulko.com> wrote:
> 
> > On Fri, Jan 07, 2022 at 09:27:05AM -0800, Tim Harvey wrote:
> > > On Thu, Jan 6, 2022 at 12:27 PM Tim Harvey <tharvey@gateworks.com>
> > wrote:
> > > >
> > > > On Thu, Jan 6, 2022 at 11:18 AM ZHIZHIKIN Andrey
> > > > <andrey.zhizhikin@leica-geosystems.com> wrote:
> > > > >
> > > > > Hello Tom,
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: U-Boot <u-boot-bounces@lists.denx.de> On Behalf Of Tom Rini
> > > > > > Sent: Thursday, January 6, 2022 7:52 PM
> > > > > > To: u-boot@lists.denx.de
> > > > > > Cc: Tim Harvey <tharvey@gateworks.com>
> > > > > > Subject: [PATCH] Revert "tree: imx: remove old fit generator
> > script"
> > > > > >
> > > > > > This reverts commit d9a6f0eed66a39206b13513ec914f14084c3bb73.
> > > > > >
> > > > > > For right now, it's too close to the release to merge the series
> > that
> > > > > > allows for binman to be used to generate the final images, and
> > also not
> > > > > > break CI, and then also merge all of the series that convert
> > currently
> > > > > > broken platforms to use binman instead.  So, bring back this
> > script now
> > > > > > and remove it again for real after the release.
> > > > >
> > > > > Please note that this might not work, as the FIT generator script
> > would
> > > > > generate ITS with '@' symbols which are not compatible with mkimage
> > due
> > > > > to CVE-2021-27138. This revert should be complemented with the fix to
> > > > > remove those '@' symbols as well.
> > > >
> > > > Correct, the revert is not enough anymore:
> > > >   MKIMAGE u-boot.itb
> > > > u-boot.its:7.11-15.5: Warning (unit_address_vs_reg): /images/uboot@1:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:16.9-21.5: Warning (unit_address_vs_reg): /images/fdt@1:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:22.9-27.5: Warning (unit_address_vs_reg): /images/fdt@2:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:28.9-33.5: Warning (unit_address_vs_reg): /images/fdt@3:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:34.9-39.5: Warning (unit_address_vs_reg): /images/fdt@4:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:40.9-45.5: Warning (unit_address_vs_reg): /images/fdt@5:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:46.9-55.5: Warning (unit_address_vs_reg): /images/atf@1:
> > > > node has a unit name, but no reg property
> > > > u-boot.its:60.12-65.5: Warning (unit_address_vs_reg):
> > > > /configurations/config@1: node has a unit name, but no reg property
> > > > u-boot.its:66.12-71.5: Warning (unit_address_vs_reg):
> > > > /configurations/config@2: node has a unit name, but no reg property
> > > > u-boot.its:72.12-77.5: Warning (unit_address_vs_reg):
> > > > /configurations/config@3: node has a unit name, but no reg property
> > > > u-boot.its:78.12-83.5: Warning (unit_address_vs_reg):
> > > > /configurations/config@4: node has a unit name, but no reg property
> > > > u-boot.its:84.12-89.5: Warning (unit_address_vs_reg):
> > > > /configurations/config@5: node has a unit name, but no reg property
> > > > ./tools/mkimage: verify_header failed for FIT Image support with exit
> > code 1
> > > > Makefile:1433: recipe for target 'u-boot.itb' failed
> > > > make: *** [u-boot.itb] Error 1
> > > > make: *** Deleting file 'u-boot.itb'
> > > > make: *** Waiting for unfinished jobs....
> > > >
> > > > I don't know what had changed to cause this or when (again, I stopped
> > > > worrying about it because I thought we were moving to binman for this
> > > > release). There was a patch that resolved this from Oliver at
> > > > https://lists.denx.de/pipermail/u-boot/2021-August/457997.html but I
> > > > don't think that fully solves anything 'at this point' either.
> > > >
> > > > Even with that applied to current master I then end up with:
> > > >   MKIMAGE flash.bin
> > > > ./tools/mkimage: Can't open spl/u-boot-spl-ddr.bin: No such file or
> > directory
> > > > arch/arm/mach-imx/Makefile:167: recipe for target 'flash.bin' failed
> > > > make[1]: *** [flash.bin] Error 1
> > > > make[1]: *** Deleting file 'flash.bin'
> > > > Makefile:1526: recipe for target 'flash.bin' failed
> > > >
> > > > At some point over the past couple of months that patch resolved the
> > > > building issue when using the FIT generator but I also don't know what
> > > > else has changed that now causes that to not work.
> > > >
> > > > As Tom pointed out in another thread these build failures did not get
> > > > caught by CI apparently because CI does a 'make all' which did not
> > > > include the FIT images (that was accomplished with the 'flash.bin'
> > > > target prior to binman conversion).
> > > >
> > > > Is it too late to apply the CI fix and the pending binman conversions?
> > > >
> > > > I know that my series has been reviewed by Marcel [1] and as far as I
> > > > know didn't get merged simply because of the CI issue. It still
> > > > applies and produces a valid flash.bin image.
> > > > I was also able to merge Peng's series [2] which converts
> > > > imx8mq_evk/imx8mq_phanbell/pico-imx8mq to binman and was able to build
> > > > flash.bin images for them
> > > >
> > > > I tried to merge Adam's series that moves imx8mm_beacon to binman [3]
> > > > and imx8mn_beacon to binman [4] but they no longer apply due to
> > > > defconfig/Kconfig changes
> > > >
> > > > That still leaves the following unbuildable with
> > > > CONFIG_SPL_FIT_GENERATOR = "arch/arm/mach-imx/mkimage_fit_atf.sh":
> > > >
> > configs/cgtqmx8_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8mm-icore-mx8mm-ctouch2_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8mm-icore-mx8mm-edimm2.2_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8mm_beacon_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8mn_beacon_2g_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8mn_beacon_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > configs/imx8qm_rom7720_a1_4G_defconfig:CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh"
> > > >
> > > > Tim
> > > > [1] https://patchwork.ozlabs.org/project/uboot/list/?series=265765
> > > > [2] https://patchwork.ozlabs.org/project/uboot/list/?series=268380
> > > > [3] https://patchwork.ozlabs.org/project/uboot/list/?series=261640
> > > > [4] https://patchwork.ozlabs.org/project/uboot/list/?series=261822
> > > >
> > >
> > > Tom,
> > >
> > > I'm not familiar with the U-boot CI tool.
> >
> > It's at https://u-boot.readthedocs.io/en/latest/develop/ci_testing.html
> > (and in-tree under doc/develop/ci_testing.rst).
> >
> > > Is it a show-stopper that it
> > > does not build for boards using binman for release? From what you
> > > mentioned in another thread it was never building the flash.bin target
> > > for the boards using the FIT generator anyway.
> >
> > Yes, breaking CI is a ship-stopper.  That all of these boards were not
> > previously building the final image as part of "all" is a problem.
> >
> > So, here's what I'm at right now.  I've grabbed Heiko's patch.
> > Everything is currently visible at
> >
> > https://patchwork.ozlabs.org/bundle/trini/2022-01-07-imx8-and-binman-updates/
> > and with this, we get some boards building and complaining as expected:
> >    aarch64:  w+   imx8mn_evk
> > +(imx8mn_evk) Image 'main-section' is missing external blobs and is
> > non-functional: blob-ext@1
> > blob-ext@2 blob-ext@3 blob-ext@4
> > +(imx8mn_evk) Image 'main-section' is missing external blobs and is
> > non-functional: blob-ext
> > +(imx8mn_evk)
> > +(imx8mn_evk) Some images are invalid
> >
> > But others are:
> >    aarch64:  +   imx8mn_beacon_2g
> > +(imx8mn_beacon_2g) ===================== WARNING ======================
> > +(imx8mn_beacon_2g) This board uses CONFIG_SPL_FIT_GENERATOR. Please
> > migrate
> > +(imx8mn_beacon_2g) to binman instead, to avoid the proliferation of
> > +(imx8mn_beacon_2g) arch-specific scripts with no tests.
> > +(imx8mn_beacon_2g) ====================================================
> > +(imx8mn_beacon_2g) Image 'main-section' is missing external blobs and is
> > non-functional: blob-
> > ext@1 blob-ext@2 blob-ext@3 blob-ext@4
> > +(imx8mn_beacon_2g) binman: Error 1 running 'mkimage -d
> > ./mkimage.spl.mkimage -n spl/u-boot-spl
> > .cfgout -T imx8mimage -e 0x912000 ./mkimage-out.spl.mkimage':
> > spl/u-boot-spl-ddr.bin: Can't ope
> > n: No such file or directory
> > +(imx8mn_beacon_2g)
> > +(imx8mn_beacon_2g) make[1]: *** [Makefile:1088: all] Error 1
> > +(imx8mn_beacon_2g) make: *** [Makefile:177: sub-make] Error 2
> >
> 
> The beacon boards are mine.  I can work on this one today.  Do I just grab
> the binman updates and apply it to master and fix it from that starting
> point?

So, yes, grabbing the series from the above link and applying it on top
of master, and making builds work with fake binaries would be very
helpful.  As best I can tell there's still something missing with making
fake blobs link and not fail.  For example, imx8mm_venice is part of the
series above but still fails with 'make BINMAN_FAKE_EXT_BLOBS=1 ...':
Image 'main-section:u-boot-spl-ddr' has faked external blobs and is non-functional: lpddr4_pmu_train_1d_imem.bin lpddr4_pmu_train_1d_dmem.bin lpddr4_pmu_train_2d_imem.bin lpddr4_pmu_train_2d_dmem.bin
Image 'main-section' is missing external blobs and is non-functional: atf_blob blob-ext
Wrote map file './imx-boot.map' to show errors
binman: Node '/binman/imx-boot/blob-ext@1': Offset 0x0 (0) overlaps with previous entry '/binman/imx-boot/uboot' ending at 0x1aee08 (1764872)

And that last error is the real problem.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]