Vasily Khoruzhick <anarsoul@xxxxxxxxx> wrote: > > On Fri, Jan 3, 2025 at 1:13 AM Ping-Ke Shih <pkshih@xxxxxxxxxxx> wrote: > > > > Vasily Khoruzhick <anarsoul@xxxxxxxxx> wrote: > > > Fix 3 typos in 8703b driver. 2 typos in calibration routines are not > > > fatal and do not seem to have any impact, just fix them to match vendor > > > driver. > > > > Just curious how you can find these typos? > > I added traces to sdio_* functions in linux (see [1]), so I can > capture register access traces. I captured the traces from both rtw88 > and the vendor driver and wrote a simple parser that decodes the > traces, see [2]. I guess it would be easier with an USB device, where > we have usbmon. I really wish there was something like usbmon for > SDIO. > > I also added traces for C2H messages to both drivers, since they go > through sdio_memcpy_fromio() that I don't trace. > > Once I had the traces, I manually compared them (along with register > dumps, rtw88 has it in debugfs, vendor driver in proc) trying to find > the writes that do not match. Unfortunately, rtw88 and vendor driver > flows are different enough, so I couldn't come up with a way to > compare it automatically > > Adrian and Bitterblue supported me on #linux-wireless on IRC, and one > of the typos in IQK calibration was actually found by Bitterblue. > > It took ~5 evenings and 1 weekend to get to REG_OFDM0_TX_PSD_NOISE > (0xce4). Once I changed it from 0 to 0x10000000 via reg_write over > debugfs, it magically fixed the issue. I changed it back to 0 to > confirm that it breaks it again, and then back to 0x10000000 to see it > working. Then it was just a matter of grep to find where this register > is written in rtw88 and compare the corresponding code to the vendor > driver. > > [1] https://github.com/anarsoul/rtl8723cs-re/blob/master/sdio_traces.patch > [2] https://github.com/anarsoul/rtl8723cs-re > Cool. I did similar works when we rewrite rtw89 from vendor drivers. Sometimes we also found flaws of vendor driver by their difference. Ping-Ke
