14:40 <shorne> somlo: FYI, this is the patch to fix LiteETH for big endian machines like openrisc: https://github.com/stffrdhrn/linux/commit/0223cdd5b77a4f05b2824c1bf6493c04ee0ad76e

14:40 <tpb> Title: Litex: LiteETH: Fixups for compat with BE, also properly detect unset… · stffrdhrn/linux@0223cdd · GitHub (at github.com)

14:41 <shorne> I am not sure if you saw the LKML mails/messages. I am planning to upstream via my openrisc tree a few of the litex drivers in the 5.10 merge window.

14:42 <shorne> those are here: https://github.com/stffrdhrn/linux/commits/or1k-5.10-litex

14:42 <tpb> Title: Commits · stffrdhrn/linux · GitHub (at github.com)

14:43 <shorne> Now, that liteETH is working I can do some work.

15:29 <somlo> shorne: afaik, the stuff being pushed on lkml is 32-bit cpu and 8-bit csr-data-width specific, and I promised Mateusz to shut up and let it happen, and start making noise about 64-bit CPUs and 32-bit csr-data-width only after the initial upstreaming pass :)

15:32 <somlo> shorne: I'll test your patch with Rocket (64-bit LE); if it doesn't break that, I'll post it to the litex-rocket-rebase branch (which in spite of its name *should* be backward compatible to 32-bit CPUs and whatever csr data width litex is built with)

15:45 <somlo> shorne: so the rocket branch I maintain already has the accessors (ioread/iowrite) updated to work across widths and endianness

15:46 <somlo> so all that's different in your version is the `readx_poll_timeout_atomic` thing, and some of the logic around `of_get_mac_address()`

15:47 <somlo> I'll apply those changes manually (as your patch is not itself based on litex-*-rebase in litex-hub, as far as I can tell)

15:49 <_florent_> Hi shorne, nice to see you here and great to see you are testing/helping with the linux drivers.

15:49 <_florent_> If you think the dts generation is missing some features (like the IRQ you reported), feel free to create issue on LiteX repo for that, i'll try to implement it

15:52 <_florent_> Otherwise, as you are probably already aware, Mateusz and somlo are working on the drivers, but if have issues with the FPGA or trouble understanding how it works, feel free to ping me, i would be happy to help and do some tests

15:57 <somlo> shorne: https://imgur.com/a/HEGz3CT -- is `inreg8` an or1k-ism ? Any way we can make this line portable across everything? https://github.com/stffrdhrn/linux/commit/0223cdd5b77a4f05b2824c1bf6493c04ee0ad76e#diff-9ec47ad4f74cb589e1024ed04c60670dR245

15:58 <tpb> Title: Imgur: The magic of the Internet (at imgur.com)

16:06 <somlo> shorne: the of_get_mac_address() thing works perfectly, and fixes an annoyance I had commented out and was hoping to return to one day, but never did -- so thanks for that! :)

16:12 <somlo> oh no, inreg8 is the silly liteeth-driver-specific accessor wrapper I thought I'd gotten rid of :)

16:16 <somlo> it's `litex_reg_readb` in the litex-rocket-rebase branch, and seems to work fine when I test it on rocket

16:42 <somlo> shorne: https://github.com/litex-hub/linux/commit/35b0a4c900cbd3c0176306043b54c82539942d57 -- thanks!

16:42 <tpb> Title: Litex: LiteETH: Fixups for compat with BE, also properly detect unset… · litex-hub/linux@35b0a4c · GitHub (at github.com)

18:16 <somlo> _florent_: https://github.com/enjoy-digital/litex/commit/e28e808c244ce1d57146c92a270ce4d08c9967cc#diff-051c0188ba9860fa5fa5242889764189R34

18:16 <tpb> Title: software/liblitedram: allow forced write leveling delays, improve del… · enjoy-digital/litex@e28e808 · GitHub (at github.com)

18:17 <somlo> it results in a build error (probably the `int sdrwl_delays[16];` array should be in sdram.c, not sdram.h (from where it gets included in multiple places, and ends up conflicting with itself)

18:29 <somlo> and ouch, the entire memspeed thing once again violently assumes that a pointer is "unsigned int" :(

18:48 <somlo> _florent_: https://github.com/enjoy-digital/litex/pull/649

18:48 <tpb> Title: misc. fixes for a few recent commits by gsomlo · Pull Request #649 · enjoy-digital/litex · GitHub (at github.com)

19:15 <_florent_> somlo: thanks for the PR

19:27 <somlo> Thanks for applying! (I'm the 64-bit CI testing bot, but I catch the occasional unrelated compile warning or error ;)

19:37 <somlo> _florent_: the memspeed/memtest thing turned out to be a lot less of a problem than I initially thought (there's lots of `unsigned int *` that's fine, only a few instances of `(unsigned int)addr` that needed to be fixed

19:37 <somlo> so I ended up more freaked out about it than I needed to be :D

20:27 <shorne> somlo: inreg8 is/was a helper in the liteeth8 driver. Let me try with your latest driver. The problem with using readb_.. I think is that readb is not for iomem, so for me the read returned nothing

20:29 <shorne> _florent_: thanks, I will let you know, I have been able to get by reading the code on my own so far. I'll report the missing interrupts

21:15 <somlo> shorne: I remembered getting rid of local-to-the-driver accessors such as inreg8 :) I think the "adapted" version of your patch in litex-rocket-rebase should work (the accessors there should be endianness and csr-data-width agnostic)

22:19 <shorne> somlo: yeah, I figured they had to be removed, but I couldn't find the branch that had any newer liteeth driver. Now I have your's thanks