00:01 <shadows> hi all, looking for someone with a VisionFive JH7100 (not the JH7110 "VisionFive 2"), discontinued product so I can't get one myself to test the config change I proposed for Debian Trixie release to get SDIO WiFi+BT AP6236 chipset working... needs testing, anyone?

00:02 <shadows> also related is the Milk-V Mars CM and Mars CM Lite which have AP6256 chipset, same config change but then additional things need done on top of this for upstream; also seeking to collaborate with anyone interested in that

00:09 <shadows> oh and thirdly I was sent a StarPro64 (EIC7700X CPU) rev1 dev board - anyone else here happen to have one of these? I'd like to learn how to enter USB Flash mask rom loader mode when there's no MSEL-DIP part on that revision of the board for boot selection

02:35 <ganboing> shadows: I have a spare JH7100. If you need one I can probably sent it to you, but let me double check when I get home.

02:39 <ganboing> I also have EIC7700X, but that's a Hifive Premier P550. I'm surprised if there's no DIP switch to boot selection.

02:40 <ganboing> From pine64's docs, there is boot sel: https://pine64.sfo3.cdn.digitaloceanspaces.com/doc/starpro64/StarPro64_schematic_v2.0-20250310.pdf

02:41 <ganboing> Oh, are you saying you have a pre-production board?

06:36 <shadows> ganboing: yes, pre-production rev1 StarPro64 that the info team sent to me. I asked about the v1 schematic and the answer was it would be confusing to customers so they're not going to publish the schematic.

06:38 <shadows> ganboing: the short question I have is about how Debian Trixie works on the JH7100, I have asked for CONFIG_BRCMFMAC_SDIO kernel option to be included with Trixie release on the guess that it may help with the SDIO WiFi+BT module functionality

06:39 <shadows> I would like to know, does a standard install of Debian Trixie (from Debian official, not any vendor thing) fail to activate the SDIO WiFi+BT module?

06:44 <shadows> what I can see is that it would require devicetree description (yes it has this), Firmware descriptions (package firmware-brcm80211 from non-free-firmware official package repository), and kernel driver support which I think is missing at least that Kconfig option

08:14 <ganboing> Which image I should be using for Debian Trixie?

08:15 <shadows> just use the net installer, at https://d-i.debian.org/daily-images/riscv64/daily/netboot/mini.efi

08:15 <shadows> something like interrupt u-boot, transfer that file to memory address $loadaddr for example, then 'bootefi $loadaddr'

08:17 <shadows> err... does that board have U-Boot bootloader? I don't know

08:41 <shadows> maybe that board belongs in a computer history museum

09:19 <dramforever[m]1> shadows: do you happen to have a link or something that i can forward off of irc/matrix

09:27 <ganboing> shadows: It looks like Debian Trixie 6.12.27-riscv64 kernel is not working well on my jh7110/vf1.

09:28 <ganboing> I booted up the installer, and the mmc/gmac is not working properly.

09:29 <ganboing> *jh7110 -> jh7100

09:29 <ganboing> I suspect it's due to the Incoherent cache in jh7100:

09:29 <ganboing> https://github.com/starfive-tech/JH7100_Docs/blob/main/JH7100%20Cache%20Coherence%20V1.0.pdf

09:32 <ganboing> The RISCV_DMA_NONCOHERENT for jh7100 is there in v6.12, not sure if it's really working properly

09:32 <ganboing> Probably not waste too much time on this soc/board

09:50 <shadows> ganboing: ah, I agree not to waste too much time on that. Thanks for trying.

09:54 <shadows> dramforever[m]1: link about the CONFIG_BRCMFMAC_SDIO ? https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1108215 and https://salsa.debian.org/kernel-team/linux/-/commit/67ed0959482e3beb6e38a6675689c6f2c2552609

09:57 <Esmil> ganboing: the RISCV_DMA_NONCOHERENT upsteram works fine for the jh7100, but you need the ERRATA_STARFIVE_JH7100 that depends on NONPORTABLE, so it can't be combined with support for many other boards

09:59 <ganboing> That's why! So I assume each distro needs to have a separate kernel package for these RISCV_DMA_NONCOHERENT boards, right?

09:59 <ganboing> Does Debian do that?

09:59 <Esmil> however the ESWIN EIC7700 has the same issue as the JH7100 and Samuel posted an RFC for a way to enable those chips without the need for the DMA_GLOBAL_POOL that needs NONPORTABLE

10:00 <Esmil> if that gets merged distributions will be able to support both the JH7100, EIC7700 and other boards in the same kernel

10:00 <Esmil> that hasn't happened yet though

10:01 <ganboing> Yea. I'm really annoyed that EIC770x is RISCV_DMA_NONCOHERENT. We already had this shit with starfive jh7100->jh7110

10:01 <ganboing> And we have to go through the same shit again...

10:08 <shadows> trying here to get the JH7110 SoM (Milk-V Mars CM) to go with its AP6256 SDIO WiFi+BT module for upstream patch to LKML and I was hopeful to have some example of the same driver and upstream devicetree configuration working as a proof that such thing is useful to extend for this one

10:18 <shadows> the cache noncoherence is a rabbithole I don't want to follow, either

10:19 <dramforever[m]1> what's the difference between this and dma-noncoherent in the dt

10:19 <dramforever[m]1> oh, no cache management instructions?

10:20 <dramforever[m]1> ... eic770x doesn't have CMO? come on

10:25 <ganboing> Yep, and no svpbmt

10:26 <shadows> do you know about dt and setting GPIO states? schematic at https://github.com/milkv-mars/mars-files/blob/main/Mars-CM_Hardware_Schematices/Milk-V_Mars-CM_SCH_V1.01_2024-0425_eMMC.pdf page #16 "WIFI/BT-SDIO" I am also looking at datasheet http://files.pine64.org/doc/datasheet/PinebookPro/AP6256%20datasheet_V1.7_12282018.pdf

10:31 <shadows> I get confused why JH7110 in mainline Linux has both PINMUX() and GPIOMUX() macros; In this situation I want to set some GPIO pads for input and some for output to work with the AP6256 WIFI/BT-SDIO chip (brcm80211 brcmfmac driver in mainline Documentation/devicetree/bindings/net/wireless/brcm,bcm4329-fmac.yaml)

10:59 <shadows> pin 12 of AP6256 is pulled down by 10k resistor on connection to WIFI_REG_ON_H_GPIO33 so what, I have pinmux = <GPIOMUX(33, GPOUT_LOW, GPOEN_ENABLE, GPI_NONE)>; along with 'bias-disable' (because external pull-down) and 'input-disable' ?

11:02 <shadows> doesn't seem to be a GPOUT_NONE defined so I am guessing that for output operation to feed that macro GPOUT_LOW (?)

11:07 <ganboing> PINMUX() is a short version of GPIOMUX()

11:08 <shadows> I did not see where in GPIOMUX define that the bits for the function select are set; aren't they distinct?

11:09 <ganboing> It doesn't. The func is 0 in GPIOMUX macro expansion

11:10 <ganboing> PINMUX can set func, but it then sets doen=1 and then dout/din=0.

11:10 <ganboing> I see PINMUX is only used for those GPIOs whose id >= 64

11:10 <ganboing> They are special GPIO's that doesn't have doen/dout, and the code treats them specially

11:15 <shadows> the vendor (Milk-V Mars git repo for Mars CM branch) code circa Linux 5.10 has dt that shuffles bit values into vendor properties ref. https://github.com/milkv-mars/mars-buildroot-sdk/blob/dev-mars-cm/linux/arch/riscv/boot/dts/starfive/jh7110-milkv-mars-cm-emmc.dts

11:17 <shadows> assuming the then GPIO_IE(1) refers to input enable true (it is setting a bit value so I'm not 100% sure that it's not inverted) it's just setting input enable true on basically everything whether it's an input or an output ?

11:23 <ganboing> I don't think it means input enable. It's very likely Interrupt Enable, because Interrupt Enable (IE), Pull-Up (PU) and Drive-Strength (DS) are in the same register

11:24 <shadows> oh I might be mis-reading this... O_O

11:25 <ganboing> Use this https://doc-en.rvspace.org/JH7110/PDF/JH7110_TRM_StarFive_Preliminary_V2.pdf section 2.7, 2.8

11:31 <shadows> the TRM uses "Input Enable" language and the description is in agreement

11:34 <shadows> Input Enable ("1: Enable the receiver"), Output Drive Strength [2mA=0,4mA=1,8mA=2,12mA=3], Pull-Up ("1: Yes"), Pull-Down ("1: Yes"), Slew Rate Control ("1: Fast"), schmitt trigger select ("1: Enable"), power-on-start ("1: Enable active pull-down for loss of core power"), ... reserved

11:34 <ganboing> My bad. Yea, it's Input Enable

11:38 <shadows> on the JH7110 if you're not using one of the defined assignments though, what exactly is going on? or maybe that's not a well formed question what I am trying to ask is about the macros and setting registers, if I'm not trying to use one of the pre-defined GPIO matrix assignments for some on-chip function and I just want enable input or output, for the purposes of the macro and bits are GPOUT_LOW and

11:38 <shadows> GPOUT_HIGH kind of virtual sources that set initial output state?

11:40 <shadows> that weird maybe-typo in arch/riscv/boot/dts/starfive/jh7110-pinfunc.h "sys_iomux gin" where elsewhere it's referred to as 'din'

11:42 <shadows> I'm looking at GPOUT_LOW GPOUT_HIGH GPOEN_ENABLE GPOEN_DISABLE and I supposed GPI_NONE then

11:44 <shadows> so, what, GPIOMUX(padnum, GPOUT_LOW for default low or GPOUT_HIGH for default high, GPOEN_ENABLE, GPI_NONE) as an output? and I guess maybe GPIOMUX(padnum, ?, ?, ?) what for input

11:45 <ganboing> Is this helpful? https://lore.kernel.org/all/20231130151932.729708-9-emil.renner.berthing@canonical.com/

11:45 <ganboing> jh7100 and jh7110 share similar GPIO/pinctrl controller

11:46 <shadows> it is relevant yeah, pretty similar

11:48 <shadows> that patch does not configure the pin used on the pwrseq node

11:48 <shadows> the mmc related pins make sense to me to some extent because all of the macro parameters are obviously what they're supposed to be

11:50 <shadows> looking at the AP6236 (for jh7100-common.dtsi) or what I have in-hand the AP6256 (Milk-V Mars CM what I'm writing dt for) it's not obvious to me how to configure those pins which connect the non-mmc lines

11:52 <shadows> at this point I don't even get Linux trying to load the brcmfmac driver and the message is that the mmc1 non-removable card could not be initialized. Huh

12:05 <ganboing> mmc1? shouldn't it be mmc0?

12:06 <shadows> on JH7110 reference board VisionFive2 the eMMC is mmc0 and SD Card (SDIO) mmc1

12:07 <shadows> for Milk-V Mars CM Lite (no eMMC) the mmc0 is configured for SD Card and that also Mars CM (with eMMC) then mmc1 is reserved for the optional AP6256 module

12:07 <ganboing> Got it. It's flipped

12:37 <ganboing> shadows: pinning you in private chat...

21:18 <palmer> I'm going to miss the Linux patchwork meeting next week, my parents and my wife's parents come to town that day so it's just going to be chaos

22:43 <drewfustini> Thanks for heads up, I hope you survive :)

22:47 <shadows> what does it mean in devicetree /soc/i2c@10030000/interrupts=<35> is that 35 a GPIO? or unrelated.

22:47 <shadows> there's so much I don't even know what I don't know