02:56 <tuxd3v> LordKalma, my bad, I had a bug in my return type for bswap64..you were right, it is the same result :)

06:27 <junari> Has anyone met devices with swapped SD and sdio?

06:29 <junari> I think it's a huge security hole that the device can boot seamlessly from any microsd card

06:31 <junari> But it will probably be easier to add a gpio controlled regulator

07:01 <szemzoa> apritzel: sure, I can help with the testing, if you give me enough time :)

10:21 <LordKalma> apritzel, I have the binaries for buildroot 2023.01 for our device with gcc 10 (working) and 12 (not working), if you want them

10:22 <apritzel> LordKalma: well, "want" is not the right word ;-) I remember from the H616 investigation that it's a pain to compare the disassemblies, and it wasn't conclusive the last time

10:23 <apritzel> LordKalma: what's more promising is to try to bisect on the function level: I mix and matched functions from both compilers back then

10:24 <apritzel> but that requires testing on the device

10:24 <apritzel> LordKalma: so sure, put them somewhere, but I cannot promise that I will find time and motivation to look at them :-(

10:27 <LordKalma> I do have en elf with debugging symbols because my colleague is really nice

10:56 <jernej> if anyone wants overengineered calculator based on A50: https://www.cnx-software.com/2023/03/02/ghlbd-android-calculator-mini-review-an-allwinner-a50-based-android-9-0-calculator/

12:26 <junari> New allwinner roadmap https://disk.yandex.ru/d/g4f0KE6Yui3sFw

12:37 <wens> big cores in late 2023 or 2024, hmm

12:42 <gamiee> very interesting

12:42 <gamiee> but well, Allwinner showed a lot of roadmaps, and only few of chips were released at all, mostly after the promised date of release.

12:42 <LordKalma> the silicon crisis is hard

12:43 <gamiee> of course, but this was also in 2019

12:46 <junari> rk3588 had to wait a long time too

12:48 <gamiee> also true

12:49 <junari> It would be nice if at least some of the powerful chips came out in the time frame that they promise

12:52 <gamiee> yeah, although, I'm bit worried

12:53 <gamiee> H616 didn't had that much interest as how older SoCs had. Of course, different times, but it seems that most of open-source workload and interest of companies shifted to Rockchip

12:53 <gamiee> I would totally appreciate have successor for H3, with bugs fixed (mainly thermal stuff), but same mainline support. that would be blast.

12:54 <jernej> gamiee: I see quite some interest regarding H616 from retro community. It's dirt cheap and has decent enough GPU for that task.

12:56 <jernej> Thing is that noone from "knowledgeable" developers is interested enough to bring up support for missing drivers in mainline

12:56 <gamiee> jernej: that's good to hear! Well, maybe it's time for me, after I will finish current product with H3, to start working on H616, and contribute by writing some drivers.

12:57 <gamiee> Exactly, and if they do, they work on Rockchips. But I guess this is because of collabora and companies funding this effort?

12:59 <jernej> well, AW wasn't really on the radar of such companies

12:59 <gamiee> Well, afaik it was back in 2018 - 2019, resp. I knew that many companies used AW SoCs in their products. But almost all of them switched to Rockchips.

13:02 <gamiee> One time, we had showcase of X-Ray manufacturer in our company, and they had few boards we could test X-ray on. One of boards was using H3, and after tracing the product by labels on pcb, I find out it some smart display product from EU company.

13:03 <gamiee> Also, for example, I heard from friend they got RK3568 and T507 and T7 dev boards, they firstly started evaluating T507 due price, but after pain with SDK and lack of support from manufacturer of SoM, they tested Rockchip SoM and it was way better for them and they're using it now in product.

13:10 <karlp> I'm using allwinner at work because it's _less_ hassle than rk..

13:10 <gamiee> karlp: same here

13:10 <karlp> I had originally tried rk, but I found the tooling and environment much worse.

13:11 <gamiee> also, it is because none of SoCs supports the features I need as on H3 (or H6). And all this in mainline

13:12 <junari> gamiee: yes, myir som with t507 costs 35$, firefly som with rk3568j costs 120$

13:14 <gamiee> @junari : exactly, but sadly, T507 support on BSP is not good, and mainline support in early stages.

13:14 <rajkosto> i finally finished my RockPi5 preorder cuz why not, its pretty fast, mainline is gonna take forever tho

13:15 <gamiee> rajkosto: yeah

13:30 <karlp> mainline t507 looks goopd for ~¨€verythign except the graphics stuff?

13:30 <karlp> looks perfectly fine mainline for headless stuff, unless I'm misreading something in the tables.

13:32 <gamiee> GPU works in H616 so it should also on T507?

13:37 <junari> karlp: t507 don't have LPDDR4 dram driver and pmic at first

13:38 <junari> And requires another ATF

13:39 <karlp> ok, that's a bigger issue, not easily visible from teh "linux mainlining" page :)

13:48 <gamiee> I just hope they will not stop tapping out H3, else I'm screwed :D

13:51 <LordKalma> apritzel, ghidra says the show_dram_config functions are different

13:53 <LordKalma> https://i.gyazo.com/874f368f161bb2978881e2bd988ddfa5.png

13:57 <LordKalma> have the two disassemblies

13:57 <LordKalma> https://gist.github.com/ruilvo/bebbddfe5f3b7c2ea414a317cca0d024

15:30 <apritzel> LordKalma: see what I mean with painful? Most of the differences look like unrelated register allocation differences

15:32 <apritzel> LordKalma: can you check whether the code size is different? This was another one of my theories back then: some memory problem, either due to stack overwriting something, or other data structures getting in the way

15:39 <LordKalma> it is

15:39 <LordKalma> you can see it in the print

15:39 <LordKalma> 360 vs 352 bytes for that function

15:41 <MoeIcenowy> smaeul: maybe I make some error... sorry for this

15:42 <LordKalma> but gcc 10 is the larger one

16:11 <apritzel> LordKalma: the problem is that the issue does not need to be in the DRAM code directly, especially if it's size related