07:42 <rockosov> narmstrong: xdarklight jbrunet Hello Neil, Martin and Jerome! I hope you're doing well! Could you please take a look at this series: https://lore.kernel.org/all/20230823213630.12936-1-ddrokosov@sberdevices.ru/? Is everything okay, or is there something that needs to be fixed? Should I wait for the next merge window?

15:42 <f_> hi

16:16 <f_> let's continue......rewriting stuff

16:16 <f_> I won't stop until U-Boot SPL succeeds at relocating its stack :)

16:23 <Kwiboo> f_: Hi, hehe, I hope you can make it work :-)

16:23 <Kwiboo> I am trying to dig some more into why I need the change ushort into uint on mmc_cmd struct, will try some more different gcc versions to see if the issue is related to my env, an issue with optimization done by gcc, or a cpu/cache issue

16:23 <f_> Hope we can make it work!

16:24 <f_> Kwiboo: yeah that change looks unnecessary unless you have a b0rked compiler

16:29 <Kwiboo> Yeah, something is very strange, looking at assembly everything looks fine on calling site: mov w0,#0x33; ... strh w0,[sp, #cmd]; ... mov w0,#0x8; ... str w0,[sp, #blocksize]

16:29 <Kwiboo> but possible wrong value is picked up from stack at called function side

16:29 <f_> Could be the stack having issues ¯\_(ツ)_/¯

16:30 <f_> in that case we could use the stack address Amlogic BL2 uses

16:30 <f_> (sometimes Ghidra is really useful even if you have sources)

16:32 <Kwiboo> the value for blocksize, 8, seem to get used as cmdidx, 51, and I have noticed similar for similar part when I tried to force to use cmd 51 when the function was called with cmd 8

16:33 <Kwiboo> Yeah, tried to look at the code in Ghidra, one bad version (ushort) an a working version (uint), but optimization has it working differently, so not a good compare

16:33 <Kwiboo> will try to find a different compiler version that works with ushort

16:34 <f_> The compiler optimises for size IIRC

16:34 <f_> (unless you changed -Os to something else)

16:34 <Kwiboo> yeah, should be optimized for size, using default options

16:35 <Kwiboo> also think adding a mb() helped at one point, but need to re-test that theory

16:36 <Kwiboo> also had same issue on my sm1 board, so I do not think it is cpu issue, probably broken compiler

16:36 <f_> Perhaps.

17:39 <f_> $ aarch64-linux-gnu-gcc --version

17:39 <f_> aarch64-linux-gnu-gcc (GCC) 13.1.0

17:39 <f_> ^ Kwiboo

20:13 <Kwiboo> f_: Have tested multiple gcc versions now, from 8.4.0-3ubuntu1 up to 13.1.0-2ubuntu2~23.04, and I get same issue with all of them

20:14 <Kwiboo> Only thing that have helped so far, is 1. ushort -> uint, 2. mb() between cmd. and data. assignments, 3. set data. values before cmd. values