Re: arm64: dts: rockchip: Disable HS400 for mmc on rk3399-roc-pc

Hi Doug,

Am 18.11.19 um 17:08 schrieb Doug Anderson:

Hi,


On Fri, Nov 15, 2019 at 3:19 AM Heiko Stübner [off-list ref] wrote:

quoted

Hi Markus,

Am Freitag, 15. November 2019, 11:37:58 CET schrieb Markus Reichl:

quoted

Am 14.11.19 um 14:10 schrieb Heiko Stuebner:

quoted

$subject is missing the [PATCH] prefix

will fix.

no need to resend just for this ... just to keep in mind for future patches ;-)

quoted

quoted

Am Montag, 11. November 2019, 10:51:04 CET schrieb Markus Reichl:

quoted

Working with rootfs on two 128GB mmcs on rk3399-roc-pc.

One (mmc name 128G72, one screw hole) works fine in HS400 mode.
Other (mmc name DJNB4R, firefly on pcb, two screw holes) gets lots of
mmc1: "running CQE recovery", even hangs with damaged fs,
when running under heavy load, e.g. compiling kernel.
Both run fine with HS200.

Disabling CQ with patch mmc: core: Add MMC Command Queue Support kernel parameter [0] did not help.
[0] https://gitlab.com/ayufan-repos/rock64/linux-mainline-kernel/commit/54e264154b87dfe32a8359b2726e2d5611adbaf3

I'm hoping for some input from other people in Cc but your mail headers
also referenced the drive-impendance series from Christoph [0], which
it seems we need to poke the phy maintainer again.

Did you check if changing the impedance helped (like the signal dampening
Philipp described in one of the replies there).

checked with

&emmc_phy {
+       drive-impedance-ohm = <33>;

gives no improvement:

That is sad ... I guess we really should disable hs400 then ...
that may give others more incentive to dive deeper ;-)

Just out of curiosity, is the problem with the strobe line, or with
hs400?  Have you tried using the solution from "rk3399-gru.dtsi"?
Namely:

        /*
         * Signal integrity isn't great at 200 MHz and 150 MHz (DDR) gives the
         * same (or nearly the same) performance for all eMMC that are intended
         * to be used.
         */
        assigned-clock-rates = <150000000>;

IIRC hs400 on rk3399 was a bit iffy but running at 150 MHz made it
much more reliable and still gave you 300 MB/s transfer rate (so much
better than hs200).  In reality many eMMC chips can't do > 300 MB/s
anyway.

I tried 150000000 and 100000000, but it did not help.

Gruß,
--
Markus

-Doug