Hello Tamas,
On 2024-11-12 22:05, Tamás Szűcs wrote:
On Tue, Nov 12, 2024 at 4:16 PM Dragan Simic <dsimic@xxxxxxxxxxx>
wrote:
On 2024-11-12 15:35, Tamás Szűcs wrote:
> I think it was totally fine to disable sdmmc2 at first, especially if
> it couldn’t be tested or wasn’t needed right away. From what I’ve
> seen, this board works great even at higher clock speeds than what
> rk356x-base.dtsi suggests. I don’t have access to the RK3568 errata,
> and there don’t seem to be any limits mentioned in the TRM either.
> Overall, this board is doing just fine as it is.
Sorry, I'm missing the point of mentioning some clock speeds? Any
chances, please, to clarify that a bit?
It's all about stress scenarios, right. Sustained transfer at maximum
clock, multiple SD/MMC blocks used concurrently. That kind of thing.
Different data rates forced. I hope that answers your question.
Ah, I see. Though, I don't think we should worry much about that,
although testing it all is, of course, a good thing to do.
> Regarding device tree overlays, they would be ideal for implementing
> secondary functions, such as PCIe endpoint mode for users with
> specific requirements. However, the primary functions for PCIe on the
> M2E will be root complex mode, along with SDIO host, etc. In my view,
> the hardware is well-designed and interconnected. Users have a
> reasonable expectation that these primary functions should work
> seamlessly without additional configuration, right out of the box.
That's basically what I referred to in my earlier response, and in my
previous response regarding the UART. Users would expect the
Bluetooth
part to work as well, but the error messages I mentioned look nasty,
so
perhaps something should be done about that first.
I'm not aware of any nasty error messages especially related to UART.
Well, MMC core will acknowledge when the platform part fails to
enumerate a device on sdmmc2, but there's nothing wrong with this.
It's not even an error -- certainly not a nasty one.
[ 1.799703] mmc_host mmc2: card is non-removable.
[ 1.935011] mmc_host mmc2: Bus speed (slot 0) = 375000Hz (slot req
400000Hz, actual 375000HZ div = 0)
[ 7.195009] mmc_host mmc2: Bus speed (slot 0) = 375000Hz (slot req
375000Hz, actual 375000HZ div = 0)
[ 13.029540] mmc2: Failed to initialize a non-removable card
This looks acceptable to me, but I'm now not really sure that we
should enable the sdmmc2 in the board dts. Let me explain.
As Jonas demonstrated with the WiFi+Bluetooth DT overlay, additional
DT configuration is needed to actually make an SDIO M.2 module plugged
into the ROCK 3B's M.2 slot work, so what do we actually get from
enabling the sdmmc2 in the board dts? Just some error messages in
the kernel log :) and AFAICT no additional functionality when an SDIO
M.2 module is actually plugged into the slot.
> Dragan, what did you mean by SDIO related power timing requirements?
Whenever there's an SDIO module, there's usually some required timing
of the power rails. Though, I don't know what's that like with the
non-standard M.2 SDIO modules that Radxa sells, which are intended to
be used on Radxa boards with "hybrid" M.2 slots.
Ok, I see. Not always. I can't comment on Radxa's SDIO module but I'm
sure it's reasonably standard. And so is the M.2 Key E on this board.
Actually, part of the appeal is that all standard buses are very
nicely wired up. I want everybody to be able to use them.
Yes, but getting it all wired in some way unfortunately doesn't mean
that it will all work without additional DT configuration in place,
as described above.
Also, I'm not really sure there's some strict standard for the "GPIO"
and "UART" M.2 modules, that part of the specification was/is a bit
blurry or perhaps even non-existent. It's been a while since I wrote
the M.2 aricle on English Wikipedia, :) maybe it's become defined
better in the meantime.
Once again, please use inline replying. [*]
[*] https://en.wikipedia.org/wiki/Posting_style
> On Tue, Nov 12, 2024 at 5:41 AM Dragan Simic <dsimic@xxxxxxxxxxx>
> wrote:
>>
>> Hello Jonas and Tamas,
>>
>> On 2024-11-11 20:06, Jonas Karlman wrote:
>> > On 2024-11-11 19:17, Tamás Szűcs wrote:
>> >> Enable SDIO on Radxa ROCK 3 Model B M.2 Key E. Add all supported UHS-I
>> >> rates and
>> >> enable 200 MHz maximum clock. Also, allow host wakeup via SDIO IRQ.
>> >>
>> >> Signed-off-by: Tamás Szűcs <tszucs@xxxxxxxxx>
>> >> ---
>> >> arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts | 8 +++++++-
>> >> 1 file changed, 7 insertions(+), 1 deletion(-)
>> >>
>> >> diff --git a/arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts
>> >> b/arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts
>> >> index 242af5337cdf..b7527ba418f7 100644
>> >> --- a/arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts
>> >> +++ b/arch/arm64/boot/dts/rockchip/rk3568-rock-3b.dts
>> >> @@ -688,14 +688,20 @@ &sdmmc2 {
>> >> cap-sd-highspeed;
>> >> cap-sdio-irq;
>> >> keep-power-in-suspend;
>> >> + max-frequency = <200000000>;
>> >> mmc-pwrseq = <&sdio_pwrseq>;
>> >> non-removable;
>> >> pinctrl-names = "default";
>> >> pinctrl-0 = <&sdmmc2m0_bus4 &sdmmc2m0_clk &sdmmc2m0_cmd>;
>> >> + sd-uhs-sdr12;
>> >> + sd-uhs-sdr25;
>> >> + sd-uhs-sdr50;
>> >
>> > I thought that lower speeds was implied by uhs-sdr104?
>>
>> Last time I went through the MMC drivers, they were implied. IIRC,
>> such backward mode compatibility is actually a requirement made by
>> the MMC specification.
>>
>> >> sd-uhs-sdr104;
>> >> + sd-uhs-ddr50;
>> >> vmmc-supply = <&vcc3v3_sys2>;
>> >> vqmmc-supply = <&vcc_1v8>;
>> >> - status = "disabled";
>> >> + wakeup-source;
>> >> + status = "okay";
>> >
>> > This should probably be enabled using an dt-overlay, there is no
>> > SDIO device embedded on the board and the reason I left it disabled
>> > in original board DT submission.
>>
>> Just went through the ROCK 3B schematic, version 1.51, and I think
>> there should be no need for a separate overlay, because sdmmc2 goes
>> to the M.2 slot on the board, which any user can plug an M.2 module
>> into, and the SDIO interface is kind-of self-discoverable.
>>
>> Of course, all that unless there are some horribly looking :) error
>> messages emitted to the kernel log when nothing is actually found,
>> in which case the SDIO/MMC driers should be fixed first. Also, I'm
>> not sure what do we do with the possible SDIO-related power timing
>> requirements?
