On Thu, Jan 6, 2022 at 2:07 AM ZHIZHIKIN Andrey andrey.zhizhikin@leica-geosystems.com wrote:
Hello Tim,
-----Original Message----- From: U-Boot u-boot-bounces@lists.denx.de On Behalf Of Tim Harvey Sent: Wednesday, January 5, 2022 8:08 PM To: ZHIZHIKIN Andrey andrey.zhizhikin@leica-geosystems.com Cc: u-boot u-boot@lists.denx.de; Stefano Babic sbabic@denx.de; Fabio Estevam festevam@gmail.com; Schrempf Frieder frieder.schrempf@kontron.de; Adam Ford aford173@gmail.com; Marcel Ziswiler marcel@ziswiler.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: mkimage_fit_atf.sh: not found
On Wed, Jan 5, 2022 at 3:34 AM ZHIZHIKIN Andrey andrey.zhizhikin@leica-geosystems.com wrote:
Hello Tim,
-----Original Message----- From: U-Boot u-boot-bounces@lists.denx.de On Behalf Of Tim Harvey Sent: Tuesday, January 4, 2022 11:48 PM To: u-boot u-boot@lists.denx.de; Stefano Babic sbabic@denx.de; Fabio
Estevam
festevam@gmail.com Cc: Schrempf Frieder frieder.schrempf@kontron.de; Adam Ford aford173@gmail.com; Marcel Ziswiler marcel@ziswiler.com; Jagan Teki jagan@amarulasolutions.com Subject: mkimage_fit_atf.sh: not found
Stefano and Fabio,
I'm seeing the imx8mm_venice_defconfig target failing to build on master due to mkimage_fit_atf.sh not found: ./"arch/arm/mach-imx/mkimage_fit_atf.sh" \ arch/arm/dts/imx8mm-venice-gw71xx-0x.dtb arch/arm/dts/imx8mm-venice-gw72xx-0x.dtb arch/arm/dts/imx8mm-venice-gw73xx-0x.dtb arch/arm/dts/imx8mm-venice-gw7901.dtb arch/arm/dts/imx8mm-venice-gw7902.dtb > u-boot.its /bin/sh: 1: ./arch/arm/mach-imx/mkimage_fit_atf.sh: not found
This has been dropped in d9a6f0eed6 ("tree: imx: remove old fit generator
script")
So why was that merged when it breaks several boards that are not switched to binman because of the CI issue?
Because the FIT generator script has been broken after commit 3f04db891a ("image: Check for unit addresses in FITs"), which has covered CVE-2021-27138. You can see the reasoning of merging d9a6f0eed6 ("tree: imx: remove old fit generator script") in the commit message.
In addition, d9a6f0eed6 ("tree: imx: remove old fit generator script") has been introduced as a part of discussion stemmed from [1], where it has been pointed out that certain boards are still using old FIT generator which is not working, and I've listed those board config files for maintainers to react.
Andrey,
I agree removing the old fit generator is the right path but I am very surprised it was merged when it broke boards that had not moved to binman yet. There has been a warning to migrate to binman but there was never a deadline.
Binman CI missing binaries verification came later I believe, that is the reason we are seeing those conversions pending. But this is unrelated to the FIT generator script removal, which was broken anyways.
I have not followed the discussion about what was wrong with the FIT generator or what was specifically 'broken' but boards at least built and booted with it and now they do not build. I do know there was a time where I needed a patch that dealt with '@' symbols and I'm not sure if that ever got merged before the FIT generator was simply removed. Again, I hadn't thought much of it because I had binman conversion patches in flight not realizing they would get stuck because of CI.
I don't really know anything about U-Boot CI but I'm also surprised it doesn't point out the fact that several boards won't currently build.
As far as I can tell the other boards that are still using SPL_FIT_GENERATOR also fail due to this (ie imx8mm_beacon_defconfig, imx8mq_evk_defconfig, imx8mm-icore-mx8mm-edimm2.2_defconfig, etc).
imx8mq_evk is already converted and I've sent a patch for it, see [1].
What is the state of the binman conversion? I submitted a series to convert my boards to binman and it has just been sitting without any response for months now [1].
I believe that the reason for your series sitting in the queue is the same as for imx8mq_evk: missing binary blobs (ATF and DDR) are failing CI builds.
Right, so imx8mq_evk (and others) are completely broken for the pending release correct?
That is correct, and there are patches to address this.
There is also a patch series from Heiko to address the "missing binary" CI issue, see [2].
Again personally I did not pay attention to d9a6f0eed6 ("tree: imx: remove old fit generator script") because I already had submitted a series to move my board to binman (other boards still had not done this however and would be broken without a change). I see that Stefano applied d9a6f0eed6 ("tree: imx: remove old fit generator script") on Oct 7 and he must not have realized it broke boards that had not moved to binman. Several of us have submitted patches to move to binman and thus also likely did not notice but those patches never got merged because of this CI issue.
I see Heiko has a v4 of his patch here https://patchwork.ozlabs.org/project/uboot/list/?series=279595. Can that get merged?
It looks to me like the following boards would be currently broken as they all use CONFIG_SPL_FIT_GENERATOR="arch/arm/mach-imx/mkimage_fit_atf.sh": cgtqmx8_defconfig imx8mm-icore-mx8mm-ctouch2_defconfig imx8mm-icore-mx8mm-edimm2.2_defconfig imx8mm_beacon_defconfig imx8mm_venice_defconfig imx8mn_beacon_2g_defconfig imx8mn_beacon_defconfig imx8mq_evk_defconfig imx8mq_phanbell_defconfig imx8qm_rom7720_a1_4G_defconfig pico-imx8mq_defconfig
I don't know exactly how the U-Boot release schedule works and when 2021.01 is released but in order to save these boards from being broken the following would have to occur: apply Heiko's [PATCH v4] binman: add support for creating dummy files for external blobs (https://patchwork.ozlabs.org/project/uboot/list/?series=279595) apply the conversion to binman series that has been pending because of the CI breakage for these boards: - Adam Ford: imx8mm_beacon* boards - Tim Harvey: imx8mm_venice boards - Andrey Zhizhikin: imx8mq_evk - Peng Fan: imx8mq_phanbell and pico-imx8mq
That would still leave the following boards broken as nobody has submitted binman patches for them. These are both maintained by Oliver Graute (added to cc): imx8qm_rom7720_a1_4G_defconfig cgtqmx8_defconfig
I believe Stefano is not currently available and I'm not sure when he is back to merge the pending binman conversion series. Additionally some of these series may need to be rebased (I know I had to do that twice because of things being moved to Kconfig)
Best regards,
Tim
Sounds like we need to revert d9a6f0eed6 ("tree: imx: remove old fit generator script")
No, because reverting this patch would bring the FIT generator which does not produce correct images due to reasons stated above.
If there are plans to revert it - then it should also be complemented with the patch that fixes the FIT generator. I'm not sure if this is a desired behavior to bring back old functionality and fix it for the sake of removing it at the later stage.
Tim
I'm not sure when the above breakage occurred but the conversion to binman resolves it and other things.
Please let me know what you expect me to do to resolve this as there is a release pending.
Best regards,
Tim [1] https://patchwork.ozlabs.org/project/uboot/patch/20211006201700.3018-1- tharvey@gateworks.com/
-- andrey
Link: [1]:
http://patchwork.ozlabs.org/project/uboot/patch/20211203161802.12699-1- andrey.zhizhikin@leica-geosystems.com/
-- andrey
Link: [1]: https://lore.kernel.org/u-boot/AM6PR06MB46912C898E0D6EA797D1EB5EA6C49@AM6PR0... Link: [2]: https://lore.kernel.org/u-boot/20220105125816.197045-1-heiko.thiery@gmail.co...