10:35 <minute> hmm, after successfull association today i have > [ 66.782759] rtw_8822cs mmc2:0001:1: failed to get tx report from firmware

10:36 <minute> and a bunch of "timed out to flush queue 1", also 2, and 0

10:36 <minute> the card can instantly recover by doing "echo ffe03000.mmc > /sys/class/mmc_host/mmc2/device/driver/unbind" and rebinding

10:36 <minute> i wonder what's the internal mechanism by which it recovers here

10:37 <minute> and if the driver could auto reset or sth

13:41 <minute> i'm experimenting a bit with rtw88_sdio.c today on a311d... what does BIT_DMA_AGG_TO_V1 mean exactly xdarklight?

13:51 <xdarklight> minute: it's the RX aggregation timeout (I don't know the unit of measure here, values have been copied from vendor drivers)

13:53 <minute> xdarklight: is it like a value that says, if there are no more packets to aggregate after this time period, flush?

14:29 <f_> I can try having a look at the MMC driver issues today

14:33 <minute> f_: cool

14:33 <f_> What's extremely weird, is in my case booting from SD card works.

14:34 <f_> Booting from eMMC is what b0rks entirely.

14:35 <f_> Actually, minute: your issues were with the eMMC as well, didn't you?

14:35 <f_> oops

14:35 <f_> your issues were with the eMMC as well, right?

14:36 <minute> f_: yes, sd is fine, emmc borked in uboot

14:36 <minute> cool that you can reproduce

14:37 <f_> Yup. Can reproduce on gxbb and gxl

14:37 <minute> :0

14:38 <f_> What was interesting to me, is that I could only reproduce while in SPL, regular U-Boot booted from eMMC just fine on gxl

14:39 <f_> Did not test booting from eMMC in regular U-Boot on gxbb simply because I have nothing on my gxbb board's eMMC

14:39 <f_> I did weird hackery in SPL to get it to load stuff from gxbb's eMMC to reproduce

14:41 <f_> Looking at my chat logs <minute> as a workaround i have a loop in mmc_init that tries up to three times <minute> not reliable

14:42 <f_> (and other stuff)

14:42 <minute> :d

14:42 <f_> Could try and see if that works for me, reliable or not

14:43 <f_> (because honestly it working a few times is better than it simply not working at all)

14:57 <minute> i think on banana pi cm4 _or_ the combo a311d+rt8822cs there is some underlying transfer issue. rx aggregation hides this issue but it becomes very pronounced when disabling rx agg. maybe just because there is a higher number of individual small transfers then and so a higher probability of running into the stall?

14:57 <minute> it is just weird to me that there are no reports of this on the net

15:13 <f_> Actually, it looks like my issue is different from yours

15:13 <minute> oh?

15:13 <f_> Or, wait a minute.

15:14 <f_> I think the behaviour was different if I used the stripped down SPL MMC driver

15:14 <f_> I might have that enabled.

15:14 <f_> > [*] Tiny MMC framework in SPL

15:15 <f_> Yup.

15:15 <f_> Let me disable that and try again.

15:16 <f_> Oh. Different error from yours.

15:16 <f_> > unable to select a mode : -74

15:17 <f_> v.s.

15:17 <f_> > unable to select a mode : -5

15:18 <f_> https://paste.debian.net/1326006/

15:23 <f_> minute: Wait, I recall there were some modifications needed in the meson_gx_mmc driver for it to happily load stuff at all

16:17 <minute> i'm trying to find in the vendor driver rtl88x2cs where they are honoring the sdio host max req size... can't find it so far

16:52 <minute> couldn't really find any pattern of why rtw8822cs has this random stalling problem

20:47 <xdarklight> minute: re: "no more packets to aggregate after this time period, flush?" - correct. if there's more incoming packets within the given period and the RX aggregation buffer still has room then that packet will be aggregated, saving host<->card communication overhead

21:28 <f_> Looking a bit more in the eMMC spec standard, and debugging...

21:31 <f_> Ok, so let's start from scratch again. It looks like it behaves weirdly when the mmc driver tries to configure the bus width..

21:37 <f_> >trying mode MMC legacy width 8 (at 25 MHz)

21:37 <f_> >trying mode MMC legacy width 4 (at 25 MHz)

21:37 <f_> >trying mode MMC legacy width 1 (at 25 MHz)

21:38 <f_> Then fails.