Skip to contentSkip to conversation

colibri_vf: CONFIG_TEXT_BASE and the Toradex eBoot flashing path

DISCUSSION Mehmet Fide 2 participants 3 messages Aug 6 – Aug 13

lore
Share & tools
lore
#0 lore raw

From: Mehmet Fide <[email protected]>

Hi,

a report rather than a patch, because I am not sure which way you want it solved.

Colibri VF50/VF61 modules that still carry the Toradex WinCE bootloader (here: "Toradex Bootloader 1.7 for Vybrid Built May 1 2020") are flashed in production by handing U-Boot to that loader:

> flashloader colibri_vf/u-boot-nand.imx Bootloader image size: 455448 bytes.
Loading done.
Flashing bootloader.
Writing 223 sector(s) of bootloader code from sector 64. Flashing completed.
> reboot

With colibri_vf_defconfig as it is today, CONFIG_TEXT_BASE=0x3f401000, the module then prints nothing at all: no banner, no character, on a console that works before and after. Building the very same tree with CONFIG_TEXT_BASE=0x3f408000, the value the Toradex 2015.04 fork used, the image boots normally and the rest of the flashing flow (NAND partitions, Linux) completes.

The difference is only where the image lands in OCRAM:

	0x3f401000: IVT entry 0x3f401000, image start 0x3f4004e8, len 0x6d000
	0x3f408000: IVT entry 0x3f408000, image start 0x3f4074e8, len 0x6d000

Our reading is that the loader is still resident when it copies the image in, so an image linked at the start of OCRAM overwrites the code doing the copy, while 0x3f408000 leaves the low 32 KB alone; we have not proved where exactly the loader keeps itself, so treat that as a guess. What is measured is the boot/no-boot difference above, on the same module, same card, same command, one build apart.

Booting the same 0x3f408000 image from NAND through the boot ROM works too, so the older base is not specific to the eBoot path.

Is the eBoot path still considered supported for these modules, in which case colibri_vf_defconfig would want to go back to 0x3f408000, or is it obsolete in favour of recovery mode, in which case it may be worth a note in doc/board/toradex? I am happy to send either patch.

Thanks,
Mehmet

#1 lore raw

Hello Mehmet

Quoted 11 lines · Mehmet Fide

From: Mehmet Fide <[email protected]>

Hi,

a report rather than a patch, because I am not sure which way you want it solved.

Colibri VF50/VF61 modules that still carry the Toradex WinCE bootloader (here: "Toradex Bootloader 1.7 for Vybrid Built May 1 2020") are flashed in production by handing U-Boot to that loader:

Quoted 7 lines

> flashloader colibri_vf/u-boot-nand.imx Bootloader image size: 455448 bytes.
Loading done.
Flashing bootloader.
Writing 223 sector(s) of bootloader code from sector 64. Flashing completed.
> reboot

what if you powercycle the board, instead of doing a reboot?

Quoted 20 lines

With colibri_vf_defconfig as it is today, CONFIG_TEXT_BASE=0x3f401000, the module then prints nothing at all: no banner, no character, on a console that works before and after. Building the very same tree with CONFIG_TEXT_BASE=0x3f408000, the value the Toradex 2015.04 fork used, the image boots normally and the rest of the flashing flow (NAND partitions, Linux) completes.

The difference is only where the image lands in OCRAM:

	0x3f401000: IVT entry 0x3f401000, image start 0x3f4004e8, len 0x6d000
	0x3f408000: IVT entry 0x3f408000, image start 0x3f4074e8, len 0x6d000

Our reading is that the loader is still resident when it copies the image in, so an image linked at the start of OCRAM overwrites the code doing the copy, while 0x3f408000 leaves the low 32 KB alone; we have not proved where exactly the loader keeps itself, so treat that as a guess. What is measured is the boot/no-boot difference above, on the same module, same card, same command, one build apart.

Quoted 7 lines

Booting the same 0x3f408000 image from NAND through the boot ROM works too, so the older base is not specific to the eBoot path.

Is the eBoot path still considered supported for these modules, in which case colibri_vf_defconfig would want to go back to 0x3f408000, or is it obsolete in favour of recovery mode, in which case it may be worth a note in doc/board/toradex? I am happy to send either patch.

I know very little on this board, eBoot and all the related, but my advise would be the following, U-Boot here should be standalone, and it should be possible to flash it from scratch on the device.

I would not focus on whatever you are getting pre-programmed on the board, apart maybe some note in the documentation.

Francesco

#2 lore raw

From: Mehmet Fide <[email protected]>

Hello Francesco,

Quoted 1 line · Francesco Dolcini

what if you powercycle the board, instead of doing a reboot?

Measured today, and it turned out not to be an eBoot problem at all. I wrote the current-defconfig image (CONFIG_TEXT_BASE=0x3f401000) into NAND from a running U-Boot and power cycled: the BootROM does not start it either. The module drops into the serial downloader, verified by the 0x23454523 UART association pattern echoing back on the console UART. Writing back an image linked at 0x3f408000 through the same procedure, the module cold boots normally.

So the eBoot observation from my report was just another face of the same thing: nothing starts an image linked at the bottom of the gfxRAM area, the low 28 KiB are not usable as a load address. It has been this way since commit c0f432c37780 ("configs: vf610: increase maximum size and enforce correct limit") in 2017; the vendor fork never took that change, which is presumably why it went unnoticed.

Quoted 2 lines

I would not focus on whatever you are getting pre-programmed on the board, apart maybe some note in the documentation.

Agreed on the standalone flashing. I have sent a two patch series: the first restores 0x3f408000 in colibri_vf_defconfig with the measurements above, the second adds the missing doc/board/toradex page for the Colibri VF50/VF61 with the build, flashing and recovery mode steps.

Thanks,
Mehmet