09:06 <mriesch> speaking of PEBKAC... trying to bring up the rk3568 mipi csi host. pd up. clocks up. reset deasserted. but readl/writel to the memory produces a hard lock on the cpu...

09:07 <mriesch> kevery: is there anything particularly strange about this ip core that needs extra consideration??

09:29 <diederik> mmind00: I still think it looks odd to see ``.cluster_reg = rk3568_vop_cluster_regs,`` in ``rk3588_vop`` and I wouldn't have rk3568 in a function name which isn't meant for just rk3566/rk3568 and it triggered my 'anomaly' alarm ... but those all point to a me-problem, hence PEBKAC

09:31 <mmind00> diederik: I agree :-P ... and yes it is very normal to re-use a function for a new soc when it provides the same functionality

09:32 <mmind00> mriesch: some other NIU clock maybe?

09:33 <mriesch> mmind00: hmm... i don't see any other clocks in downstream.. but let's grep around

09:34 <diederik> I probably should've double-checked it, but IIRC in that patch it went from sth like 'vop2' to 'rk3568' while the code (mostly?) stayed the same

09:39 <mriesch> mmind00: these NIU clocks seem to be even more mysterious than the other clocks... are they even managed by linux?

09:41 <mmind00> mriesch: "it depends" ... i.e. the NIU clocks are supplying the interconnect ports ... on older socs they mostly were just set to "critical" until "we model the interconnect", which never happened. For rk3588 sre has the linked-clock series (which I need to apply) ... rk3568 sits probably in the middle

09:41 <mmind00> NIU -> native interface unit or so

09:48 <mriesch> ok, power domain seems to work fine, i'd expect a kernel panic if the power domain was off

09:48 <mriesch> let's check the clocks then :-)

10:00 <CounterPillow> iirc the RK356x does indeed require NIU fiddling for some things, which is why some video clock stuff is currently marked critical

10:02 <CounterPillow> might also be worth dumping the GRF regs and comparing them across BSP and mainline

10:08 <mriesch> ok...

10:10 <mriesch> so maybe a clock needs to be added to rk3568_cru_critical_clocks?

10:10 <mriesch> (in drivers/clk/clk-rk3568.c)

10:10 <CounterPillow> 721bf080f249ab2adcc4337abe164230bfb8594f is one example of macromorgan marking something as critical due to NIU on that SoC

10:11 <CounterPillow> and 6931f85c29d5a0261219cf8a73773d3165806d84 is sascha hauer doing the same

10:13 <CounterPillow> (in case it wasn't clear, marking them critical is the "wrong" fix but it's the way to go until linked clocks can be modelled correctly in mainline, at which point I assume nobody will care enough anymore to fix it ;))

10:13 <mriesch> CounterPillow: gotcha! will play around with that

10:27 <CounterPillow> mriesch: looks like that pclk_vi_niu depends on hclk_vi_niu not being gated, according to 2.8.6 in the TRM

10:28 <CounterPillow> so marking hclk_vi critical might work?

10:29 <mriesch> let me try that

10:31 <mriesch> nice! that works!

10:31 <mriesch> thanks a lot, CounterPillow and mmind00

10:31 <CounterPillow> \o/

10:32 <CounterPillow> ok, now we just need the linked clocks work merged, and use it in rk3568 :D

10:32 <mriesch> or i could apply this one line hack and get away with it :-)

12:28 <naoki> mmm

12:29 <naoki> ROCK 4SE sdio wifi (brcm43455) works only when cold boot :(

12:29 <naoki> after reboot:

12:29 <naoki> [ 0.848355] mmc_host mmc2: Bus speed (slot 0) = 300000Hz (slot req 300000Hz, actual 300000HZ div = 0)

12:29 <naoki> [ 0.892465] mmc_host mmc2: Bus speed (slot 0) = 200000Hz (slot req 200000Hz, actual 200000HZ div = 0)

12:29 <naoki> [ 0.808370] mmc_host mmc2: Bus speed (slot 0) = 400000Hz (slot req 400000Hz, actual 400000HZ div = 0)

12:29 <naoki> [ 0.944842] mmc_host mmc2: Bus speed (slot 0) = 100000Hz (slot req 100000Hz, actual 100000HZ div = 0)

12:29 <naoki> [ 1.002457] mmc2: Failed to initialize a non-removable card

12:30 <naoki> ROCK 4C+ has same sdio wifi, it works fine

12:31 <naoki> ^vanilla kernel

12:39 <naoki> on OpenWrt on 4SE, mysteriously mmc(sdio) clock tuning always succeeds, but brcmfmac driver reports an error on initialization. it works fine on 4C+.

14:59 <mmind00> CounterPillow: at least step one I hope to achieve today ;-)

15:05 <mriesch> mmind00: the linked-clock thingy?

15:05 <mmind00> mriesch: yep

15:06 <mriesch> should i send the "mark hclk_vi as critical" patch then? i guess there is still some work to do for the rk3568 even if sre's series is merged

15:07 <mmind00> mriesch: yep

15:07 <mmind00> ;-)

15:07 <mmind00> to explain ... the rk3588 carries the definition for the clock links for a while already

15:08 <mmind00> "just" missing using them more intelligent ... that's the part sre's series does ... rk3568 even misses the basic definitions I think

15:08 <sre> while the linked-clk stuff itself should work for rk356x, there is definitely some work to do

15:09 <sre> apart from properly defining the NIU clocks, it is also missing some of the cleanups I did for rk3588

15:13 <sre> But it might be sensible to introduce something similar to the "RK3588_LINKED_CLK" define currently used by RK3588 clocks for RK3568.

15:16 <mriesch> ok i'll take a look it, next week or so. thx!

16:26 <diederik> mmind00: thanks for working on that :-D

20:30 <naoki> mmind00: Do you remember the ROCK 3A sdmmc CD issue? It still happens, so I want to add "broken-cd" to &sdmmc.