11:05 <arnd> apritzel, plaes: linusw mentioned to me that he may be able to get the cw1200 datasheet, but he just went on his extended summer vacation

11:41 <jernej> arnd: I managed to load FW with cw1200 driver and I get encouraging init done message, but then FW on chip crashes - presumably because XR819 API is slightly different (only few messages).

11:43 <jernej> cw1200 driver is from pre-OF time, it needs some improvements in this regard

11:45 <jernej> warpme_: DPLL value depends on source clock frequency for XR819

11:45 <jernej> on all AW boards that is 24 MHz

11:46 <jernej> maybe you have wrong config file?

11:48 <arnd> jernej: I think it's less a matter of "pre-OF times" and more of being intended as a standalone SDIO device that can be discovered without DT data

11:49 <arnd> but that is probably also the difference between testing a prototype device, and using the driver for integration into an actual machine where you end up needing additional data for whatever reason

11:49 <arnd> ah, I see the culprit now:

11:50 <arnd> drivers/net/wireless/st/cw1200/cw1200_sdio.c:static struct cw1200_platform_data_sdio *global_plat_data = &sagrad_109x_evk_platform_data;

11:50 <jernej> exactly

11:50 <arnd> there is even a comment saying "this cannot possibly work"

11:50 <jernej> there is a ton of options, most of which are imo not needed

11:51 <jernej> but clock frequency is important

11:51 <arnd> linusw also mentioned that the chip is on some of his mach-ux500 (an original st-ericsson SoC) boards, and he never got it to work there

11:51 <arnd> at the time that the driver was mainlined, you could not even pass DT data to an SDIO device at all, but that generally works fine these days

11:52 <jernej> actually config file is read later in the process and it compares if clock frequency match that of platform data, which is wrong imo

11:53 <arnd> jernej: is that config file the '.sdd_file = "sdd_sagrad_1091_1098.bin"' reference?

11:53 <jernej> yes

11:53 <arnd> ok, I see

11:53 <jernej> but only 2 or 3 things are checked from ~800 byte file

11:53 <jernej> so I would rather put that in DT, because it's board specific anyway

11:54 <jernej> and I don't like file parsing in kernel

11:54 <arnd> the other things (ref_clk, have_5ghz, reset, powerup, gpio) translate easily into DT properties, presumably in exactly the way that the xradio driver already does

11:55 <arnd> is there another file that gets loaded for the executable firmware, or is this the only file?

11:55 <jernej> no, there is also bootloader (only some chip variants need it) and FW

11:56 <arnd> ok

11:56 <arnd> I agree on the file parsing, though I'm not sure what regulatory requirements there might be, I've seen similar things in the ath9k driver

11:57 <arnd> firmware loading is hard, as the files you load may be specific to the board, the driver version or both

11:58 <arnd> because of the driver version issue, you can't just hardcode a firmware file name in the dt

11:58 <arnd> and because of the board requirement, you can't just hardcode it in the driver either

11:59 <arnd> so usually you'd end up with the driver constructing a file name based on some data it gets from the device (either register information, or DT) and things it knows about, and then attempts to load a file for that combinations, with possible fallbacks to a more generic binary

12:00 <jernej> yeah, like it's done for BRCM wifi

12:01 <jernej> but I don't think XR819 FW and bootloader are board specific

12:01 <jernej> as soon as I manage to make it work, I'll test it with two FW versions I have

12:02 <jernej> (extracted from Android images for different boards)

12:04 <jernej> bootloaders are even exactly the same for both boards and I think SDD files too (they use same clock frequency)

12:42 <arnd> I looked at the sdd parsing now, and agree with your conclusion, this code is completely bonkers

12:44 <arnd> it's just an extremely elaborate way to set a "listen_interval" setting. A lot of other drivers have a per-device setting with the same name, not sure where they get it from, or how one would determine the correct value for it otherwise

12:45 <arnd> I do see that cw1200_tx_h_bt() overrides the listen_interval again with a value from a management frame if that is smaller

12:46 <jernej> yeah, and all that if BT is even supported

12:46 <arnd> ah wait, the management frame gets the value from the SDD, not the other way round

12:48 <arnd> I suspect this value should actually come from user space through nl80211_set_station()

13:59 <warpme_> jernej: just fyi: if this might be helpful in _any way_ (i.e. registers compare, etc): i have preliminary working xr819 on 5.13 with karabek code

14:20 <jernej> warpme_: it would be helpful if you can change XRADIO_DBG_LEV_WSM and XRADIO_DBG_LEV_XR to 0xff in debug.h and provide dmesg after that

14:20 <jernej> note, it may be pretty big

14:21 <jernej> but some simple operation like ifconfig wlan0 up should be enough already, since this already fails in my case

15:07 <warpme_> jernej: as you wish: https://pastebin.com/gUFuVf0i (log from connecting to AP with XRADIO_DBG_LEV_WSM and XRADIO_DBG_LEV_XR to 0xff in debug.h)

15:09 <jernej> warpme_: thanks! that is a good reference