On 08/01/2017 21:30, Frank Kunz wrote:

Hello Andreas,

thanks for your feedback.

...

firmware="ofw" sounds wrong? The soc first stage bootloader (ROM) reads the first sector of the SD card and scans the partition table for a partition with the ID 0xa2. If this is found it loads a header + spl loader from that partition. The upstream u-boot spl configuration for that board expects the uboot also copied to that special partition and it expects that it needs to be the first partition. firmware="ofw" creates a small extra partition at the beginning of the partition table. This is used for header+spl+uboot. The uboot install script is adjusting the partition type and copies the spl with uboot into it.

...

Better check for file presence than for image name (think "sockit"). The file is not present since it is not copied by KIWIBoot.pm sub copyBootCode { ... KIWIQX::qxx ("mv $dest/boot/*.bin $target &>/dev/null"); KIWIQX::qxx ("mv $dest/boot/*.dat $target &>/dev/null"); KIWIQX::qxx ("mv $dest/boot/*.img $target &>/dev/null"); KIWIQX::qxx ("mv $dest/boot/*.imx $target &>/dev/null"); KIWIQX::qxx ("mv $dest/boot/*.elf $target &>/dev/null"); ... } The header+spl+uboot image file name is u-boot-with-spl.sfp and that pattern is not copied. Therefore I used the image name instead of the file presence. The file is copied by

cp /usr/src/packages/KIWIROOT-oem/boot/u-boot-with-spl.sfp boot/

in the uboot install script, which is just a hack. The clean solution would be a upstream patch for kiwi.

We basically moved away from copying random files magically within kiwi to explicitly moving things around inside the u-boot-install/setup scripts. It's much more maintainable that way. So if you already do it inside the u-boot-install script, that sounds like a perfectly valid solution.

...

For the gfx command line you may want to add a second console=tty or so? That did not help so far so solve my login problem. In the mean time I started to modify the initrd that the /boot/mbrid file does not match. But no idea yet what could be wrong. I haven't found out where it is set during the image build process, how to debug that?

The mbrid file gets generated by kiwi when it writes the MBR ID field in the MBR of the target device. So they really should be in sync. If they are not, something goes wrong in the u-boot-install/setup scripts. Also, have you verified that the target device is actually visible from within the kiwi initrd? Maybe the module is missing? To verify, just boot with kiwidebug=1 as kernel parameter and you should get into a shell after kiwi failed to find its boot device.

...

Does EFI not work yet or have you not tried? No, I have not tried yet. Not sure if that will work.

EFI booting makes it much easier to modify things like the kernel command line. It also allows you to install and select multiple kernel versions in parallel. To make use of it, all you need is a valid "$fdtfile" variable in the U-Boot default environment. Then set the respective define in the Images.kiwi.in file for your board and you should be set. You might need to convert the GPT that kiwi generates to an MBR. See existing snippets in the u-boot scripts for that. Alex -- To unsubscribe, e-mail: opensuse-arm+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-arm+owner@opensuse.org

On 09/01/2017 21:20, Frank Kunz wrote:

Am 08.01.2017 um 22:29 schrieb Alexander Graf:

...

...

...

For the gfx command line you may want to add a second console=tty or so? That did not help so far so solve my login problem. In the mean time I started to modify the initrd that the /boot/mbrid file does not match. But no idea yet what could be wrong. I haven't found out where it is set during the image build process, how to debug that?

The mbrid file gets generated by kiwi when it writes the MBR ID field in the MBR of the target device. So they really should be in sync. If they are not, something goes wrong in the u-boot-install/setup scripts.

Also, have you verified that the target device is actually visible from within the kiwi initrd? Maybe the module is missing?

I modified the initrd init/include scripts to print the device names that were used for the search run, and the mmcblk0 is there, and some value is read out. So the driver seems to be ok. The value read from /boot/mbrid is a different value.

That sounds to me like something is messing with the MBR after it got created. The MBR ID inside the raw image's MBR and the mbrid file are both created by kiwi and usually well aligned.

...

To verify, just boot with kiwidebug=1 as kernel parameter and you should get into a shell after kiwi failed to find its boot device.

That is the problem I have. The shell is started on /dev/tty2, but I need it on the same tty as the serial console is. How can this be configured?

You should get the terminal on the serial console if the serial console is the default output (console=ttyS0 / ttyAMA0 / whatever only, no console=tty)

...

...

...

Does EFI not work yet or have you not tried? No, I have not tried yet. Not sure if that will work.

EFI booting makes it much easier to modify things like the kernel command line. It also allows you to install and select multiple kernel versions in parallel.

To make use of it, all you need is a valid "$fdtfile" variable in the U-Boot default environment. Then set the respective define in the Images.kiwi.in file for your board and you should be set. You might need to convert the GPT that kiwi generates to an MBR. See existing snippets in the u-boot scripts for that.

That sounds interesting. I quick tested that by just setting the environment variable and by that two partitions are generated. First partition is a EFI-System partition and the second one a Linux filesystem. This cannot be booted by the soc first stage loader. It expects a partition with ID 0xa2 that has a header+spl loader. For the upstream uboot it has to be the first partition on the SD card. How to add an arbitrary first partition with KIWI oem?

Now it gets tricky. To simplify the code, U-Boot's internal scripts assume that the EFI partition is always the first partition. So we need to keep it there. What SoC exactly are you targetting? Usually most of them have some way to put the SPL at a magic offset on the SD card. Alex -- To unsubscribe, e-mail: opensuse-arm+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-arm+owner@opensuse.org

On 01/10/2017 12:51 PM, Andreas Färber wrote:

Am 10.01.2017 um 00:04 schrieb Alexander Graf:

...

On 09/01/2017 21:20, Frank Kunz wrote:

...

...

...

...

For the gfx command line you may want to add a second console=tty or so? That did not help so far so solve my login problem. In the mean time I started to modify the initrd that the /boot/mbrid file does not match. But no idea yet what could be wrong. I haven't found out where it is set during the image build process, how to debug that? The mbrid file gets generated by kiwi when it writes the MBR ID field in the MBR of the target device. So they really should be in sync. If they are not, something goes wrong in the u-boot-install/setup scripts.

Also, have you verified that the target device is actually visible from within the kiwi initrd? Maybe the module is missing? I modified the initrd init/include scripts to print the device names

Am 08.01.2017 um 22:29 schrieb Alexander Graf: that were used for the search run, and the mmcblk0 is there, and some value is read out. So the driver seems to be ok. The value read from /boot/mbrid is a different value. That sounds to me like something is messing with the MBR after it got created. He's calling parted in uboot-image-install.in:

https://build.opensuse.org/package/rdiff/home:frank_kunz:branches:openSUSE:Factory:ARM/JeOS?opackage=JeOS&oproject=openSUSE%3AFactory%3AARM&rev=19

(Any reason for that placement of the hunk instead of adding in the bottom?)

You may need to save and restore mbrid as done some lines up.

Ah, that explains the missing mbrid, yes :). I see two potential options according to the boot documentation: https://www.altera.com/content/dam/altera-www/global/en_US/pdfs/literature/a... 1) Use raw offset According to the documentation you can not use a partition table when using the raw format. Maybe you can trick around with abusing the first 440 bytes for the boot loader and leave the actual partition table and MBRID intact: https://en.wikipedia.org/wiki/Master_boot_record#Sector_layout Maybe there's some way to use Preloader 1 instead of 0? 2) Create a partition in the "hole" before the first partition MBR partitions don't have to be sorted. You can easily create a partition number 4 for example that fills 1kb-1M. That region is unused in images created by kiwi in the MBR case. If you want to use a GPT, the hole starts a bit later (I keep forgetting where - just try 128kb): https://en.wikipedia.org/wiki/GUID_Partition_Table That space should be enough to fit your boot loader (SPL as well as U-Boot) in hopefully. Keep in mind that you need to restore that special partition after the firstboot resize. The contents will still be there, you merely need to add the partition. Alex -- To unsubscribe, e-mail: opensuse-arm+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-arm+owner@opensuse.org

Am 10.01.2017 um 22:08 schrieb Frank Kunz :

...

Am 10.01.2017 um 13:13 schrieb Alexander Graf:

...

On 01/10/2017 12:51 PM, Andreas Färber wrote:

...

Am 10.01.2017 um 00:04 schrieb Alexander Graf:

...

...

>> For the gfx command line you may want to add a second console=tty >> or so? > That did not help so far so solve my login problem. In the mean > time I > started to modify the initrd that the /boot/mbrid file does not > match. > But no idea yet what could be wrong. I haven't found out where it is > set > during the image build process, how to debug that? The mbrid file gets generated by kiwi when it writes the MBR ID field in the MBR of the target device. So they really should be in sync. If they are not, something goes wrong in the u-boot-install/setup scripts.

Also, have you verified that the target device is actually visible from within the kiwi initrd? Maybe the module is missing? I modified the initrd init/include scripts to print the device names

On 09/01/2017 21:20, Frank Kunz wrote: Am 08.01.2017 um 22:29 schrieb Alexander Graf: that were used for the search run, and the mmcblk0 is there, and some value is read out. So the driver seems to be ok. The value read from /boot/mbrid is a different value. That sounds to me like something is messing with the MBR after it got created. He's calling parted in uboot-image-install.in:

https://build.opensuse.org/package/rdiff/home:frank_kunz:branches:openSUSE:Factory:ARM/JeOS?opackage=JeOS&oproject=openSUSE%3AFactory%3AARM&rev=19

(Any reason for that placement of the hunk instead of adding in the bottom?)

You may need to save and restore mbrid as done some lines up.

Ah, that explains the missing mbrid, yes :).

I see two potential options according to the boot documentation:

https://www.altera.com/content/dam/altera-www/global/en_US/pdfs/literature/a...

1) Use raw offset

According to the documentation you can not use a partition table when using the raw format. Maybe you can trick around with abusing the first 440 bytes for the boot loader and leave the actual partition table and MBRID intact:

https://en.wikipedia.org/wiki/Master_boot_record#Sector_layout

Maybe there's some way to use Preloader 1 instead of 0?

Good idea. It is possible. The rom loader first checks for a partition 0xa2. From there it tries to load the spl, where there is one and three backups. If that fails it tries to load in raw mode from sector 0 and does the same. If the loading and crc check of spl from offset 0 fails, it automatically continues at sector offset 0x80 with the first backup spl. So when copy the spl+uboot image to the disk by

dd if=u-boot-with-spl.sfp of=/dev/sdX bs=64k skip=1 seek=1

the rom loader loads the spl from preloader 1 location in raw mode and starts it. But the spl has a bug to load the attached uboot from the location 0x200 then. So I need to prepare some upstream patches for uboot to get that fixed and enabled in the defconfig (let us see if it will be accepted). Anyway the environment does not have the distro_boot variables in place, so I will upstream that as well.

By that no special partition is needed anymore then, but the "hole" between partition table and first partition needs to be big enough to store the spl images and the uboot (approx. 700kiB).

That is always the case, as parted aligns partitions to at least MB boundaries. Please double-check where the GPT contains values and that none of them are beyond 64kb. You will want to make this GPT future proof ;) Alex

Br, Frank -- To unsubscribe, e-mail: opensuse-arm+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-arm+owner@opensuse.org

-- To unsubscribe, e-mail: opensuse-arm+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-arm+owner@opensuse.org