08:40 <kuba2k2> Hi, I'm trying to run mainline U-Boot (v2021.10) from NAND flash on a Q8-format A13 tablet, and I'm running into every problem possible. Here are few things that I tried:

08:40 <kuba2k2> 1. flashing stock LiveSuit .img - Android boots, as expected (so the board is not faulty)

08:40 <kuba2k2> 2. flashing U-boot and SPL (with/without ECC) directly to NAND (as described in board/sunxi/README.nand) - board jumps to FEL, doesn't detect the image

08:41 <kuba2k2> 3. unpacking the .img, swapping boot0.bin with SPL, repacking ("dragon.exe" tool), flashing with LiveSuit - no signs of booting, no FEL, no serial output (I think it jumps somewhere and hangs)

08:41 <kuba2k2> 4. using vendor boot0, using U-Boot as boot1 (wrapped in eGON header / not wrapped / with base addr changed in Kconfig / not changed) - boot0 runs, I see serial output, then hangs at "Jump to Boot1."

08:42 <kuba2k2> 5. using vendor boot0 and boot1, putting U-Boot in FAT16 bootloader partition - all three bootloaders run sequentially, but U-Boot can't access neither NAND nor SD-Card - probably Boot1 is leaving them in an unusable state, or something (both media work when booting UB from SD or FEL)

08:42 <kuba2k2> About boot0: it seems that A13 (or this specific board) doesn't detect SPL when written directly to NAND; LiveSuit images are somehow spread into 1140-byte parts over the first 2 megabytes of the NAND (the first MB and second are equal/redundant). The data has some kind of ECC applied, as it doesn't resemble the source binary at all. I don't know if this behavior is specific to A13, but the only place that mentions this specific format is a

08:42 <kuba2k2> thread on Google Groups from 2013.

08:43 <gamiee> kuba2k2: hi, tbh, NAND never worked correctly for me

08:45 <kuba2k2> I see, but I think there has to be a way - if boot0 can run, then SPL should too

08:45 <kuba2k2> at least I would like to have any serial output from SPL on the NAND

08:46 <kuba2k2> which shouldn't be impossible, as I guess it's just a matter of getting base addresses right and jumping to the correct position

08:46 <gamiee> that's true

08:48 <gamiee> Also, I guess you tried this: https://github.com/u-boot/u-boot/blob/master/board/sunxi/README.nand

08:49 <kuba2k2> yes, that's what I described in 2.

08:50 <gamiee> ah yes, I'm blind, sry. In 2) by "doesn't detect the image" mean, that after flashing the SPL to NAND, BootROM still can't detect the SPL ?

08:52 <kuba2k2> yes, it acts as if it weren't there at all, jumps to FEL

08:52 <kuba2k2> if the NAND was erased it would do the same

08:53 <kuba2k2> I have a dump of NAND that is detected by BROM (stock boot), I can send it here if you tell me how :D

08:54 <gamiee> wait wait wait :D

08:54 <gamiee> So, when you do 2), and run "sunxi-fel exe 0x4a000000", the u-boot is not turned on?

08:54 <gamiee> So, when you do 2), and run "sunxi-fel exe 0x4a000000", the u-boot is not executed?

08:55 <kuba2k2> no, I didn't mean this - FEL can run u-boot fine, SD-card also

08:56 <kuba2k2> all "nand write" commands complete properly, "nand dump" shows the content that I just flashed

08:56 <kuba2k2> it's the part "You can now reboot and enjoy your NAND." that doesn't work

08:56 <kuba2k2> because after rebooting, all I get is FEL

08:57 <gamiee> take look on this: https://linux-sunxi.org/NAND#More_information_on_BROM_NAND

09:02 <kuba2k2> I think I read the entire wiki by now :) my NAND is 4096B/page, 224B OOB, 24b/1024B ECC. I'm not sure how should I understand the values at the bottom.. is it in "pages" or "bytes"? are they different for 4096B pages (the article mentions 2048B)

09:06 <kuba2k2> https://github.com/kuba2k2/a13-q8

09:07 <kuba2k2> here I uploaded first 2MB of the NAND after flashing stock firmware (this boots fine)

09:07 <kuba2k2> you can see how weirdly this is split in 1140-byte chunks on every page

09:09 <kuba2k2> the input SPL file is 24k in size, but on the NAND there are 256 chunks of 1140-byte each, which gives about 290k in total

09:21 <kuba2k2> For now I think flashing repacked images with LiveSuit Isa better idea, as it somehow generates that weird chunked format from standard eGON images. So it would help if I knew how to troubleshoot the SPL not running after actually being detected by BROM

09:23 <kuba2k2> Which is 3) and 4)

14:11 <kuba2k2> There has been some development in https://groups.google.com/g/linux-sunxi/c/Bpp9qaD97tg which actually describes the same issue - jumping to FEL when using "nand write" / failing SCSI operations when using boot0+boot1+u-boot / some images hanging without FEL

14:12 <kuba2k2> I will try to come up with a solution from the last email

15:50 <kuba2k2> okay, after all these days I finally have something successful! SPL image built as described in the last mail boots correctly

15:55 <libv> kuba2k2: congrats, make sure to document your findings :)

15:56 <kuba2k2> will do, after I get a working system

16:21 <kuba2k2> the thread describes additional steps about preceding Boot1 (u-boot) with 2000h bytes of zeroes, because LiveSuit refuses to flash a plain u-boot binary (it needs the eGON header). Currently, I'm trying to patch the binaries in the LiveSuit image (update_boot1.axf) to do what I tell them :)

20:55 <kuba2k2> ok, I think I partially know what can the issues be caused by:

20:57 <kuba2k2> while disassembling update_boot1.axf, I see that it's updating offset 0x81D0 of the eGON header - which corresponds to "storage data"

20:58 <kuba2k2> it fills that part with data that it receives from some parent program, I think (i.e. one of the other .axf binaries pass the data here, possibly based on script.fex)

20:59 <kuba2k2> then it *recalculates the checksum* (after first verifying the CRC before modifications), so that Boot0 thinks it's correct

20:59 <kuba2k2> after that, the modified boot1 image is written to NAND

21:00 <kuba2k2> If you take a look at the header of Boot0, "storage data" is on offset 0x1C8

21:00 <kuba2k2> that's probably why flashing the sunxi-spl.bin image from U-Boot mainline using LiveSuit just hangs the board

21:01 <kuba2k2> (because the SPL is corrupted during flashing, by the .axf files)

21:01 <kuba2k2> it just doesn't expect actual program data at that offset - for some reason, sunxi-spl starts at 0x60

21:04 <kuba2k2> in the thread on Google Groups that I mentioned earlier (and with which I had my first success booting SPL from NAND), the SPL starts at 0x2C8, through a modified mksunxiboot - right after "storage data" - that's why it's not corrupted and it works.

21:06 <kuba2k2> I can't believe it was that simple... to help other people, I will try to document all this in some github repo, and put some of it in the wiki (if I figure out how to)

21:06 <gamiee> kuba2k2: thanks for research. Putting things to wiki is simple, just create account and press edit on the page :)

21:08 <kuba2k2> Sure, thanks. With a bit more time and determination, one could possibly research the exact algorithm for putting boot0 on the NAND, in this weird chunked format... but I don't think that's me :D anyway, I'm another step towards a fully NAND-only system

22:13 <kuba2k2> so far so good, the modified binary is flashing raw U-Boot to the correct location, using LiveSuit

22:13 <kuba2k2> not so good, it doesn't boot past SPL :)