Re: [PATCH] mmc: sdhci-sprd: remove prefer asynchronous probe

[Ulf Hansson]

On Fri, 2 Dec 2022 at 03:41, Wenchao Chen <wenchao.chen666@xxxxxxxxx> wrote:
>
> On Thu, Dec 1, 2022 at 6:41 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> >
> > On Thu, 1 Dec 2022 at 10:12, Wenchao Chen <wenchao.chen@xxxxxxxxxx> wrote:
> > >
> > > The prefer asynchronous probe affects the order of device probes.
> >
> > Is there a problem with this? If so, can you elaborate on what kind of problems?
> >
>
> Hi Uffe
> dts:
> sdio0: sdio@71100000 {
> ...
> bus-width = <4>;
> no-sdio;
> no-mmc;
> ...
> }; // SD Host(Device: SD CARD)
>
> sdio3: sdio@71400000 {
> ...
> bus-width = <8>;
> non-removable;
> no-sdio;
> no-sd;
> ...
> }; // eMMC Host(Device: eMMC)
>
> According to the order of DTS: sdio0: sdio@71100000 is MMC0, sdio3:
> sdio@71400000 is MMC1.
> Log:
> [    1.714617][   T11] mmc1: SDHCI controller on 71400000.sdio
> [71400000.sdio] using ADMA 64-bit
> [    1.715276][    T9] mmc0: SDHCI controller on 71100000.sdio
> [71100000.sdio] using ADMA 64-bit
> [    1.884525][   T75] mmc1: Host Software Queue enabled
> [    1.890965][   T75] mmc1: new HS400 Enhanced strobe MMC card at address 0001
>
> But set prefer_asynchronous_probe, sometimes there will be sdio0:
> sdio@71100000 is MMC1, sdio3: sdio@71400000 is MMC0.
> Log:
> [    1.619888][   T68] mmc0: SDHCI controller on 71400000.sdio
> [71400000.sdio] using ADMA 64-bit
> [    1.620534][    T9] mmc1: SDHCI controller on 71100000.sdio
> [71100000.sdio] using ADMA 64-bit
> [    1.792570][   T77] mmc0: Host Software Queue enabled
> [    1.799532][   T77] mmc0: new HS400 Enhanced strobe MMC card at address 0001

Neither the order of the nodes in DT or dropping
PROBE_PREFER_ASYNCHRONOUS provides the guarantees that you suggest
above. In fact, that has never been the case for eMMC/SD cards.

>
> We use emmc as rootfs (without ramfs or initfs), but with SD card
> inserted and not inserted, the partition name of emmc is different, we
> need to fix this partition name.

If you can't use PARTUUID/UUID, which is certainly the preferred
solution - you may also assign a fixed index mmcN to an MMC host
controller and its corresponding mmcblk[n] devices, by defining an
alias in the /aliases device tree node.

[...]

Kind regards
Uffe