linux-kernel - Re: [PATCH 2/2] mmc: dw_mmc: use standard "mmc" alias stem

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <CAPDyKFoweEnqRWRBUsBgmApKoY30QuNp1C7f9gW5uCJbBk6A4w@mail.gmail.com>
Date:   Wed, 24 Nov 2021 16:13:54 +0100
From:   Ulf Hansson <ulf.hansson@...aro.org>
To:     John Keeping <john@...anate.com>
Cc:     Jaehoon Chung <jh80.chung@...sung.com>, linux-mmc@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH 2/2] mmc: dw_mmc: use standard "mmc" alias stem

On Wed, 24 Nov 2021 at 13:29, John Keeping <john@...anate.com> wrote:
>
> On Wed, Nov 24, 2021 at 06:54:12PM +0900, Jaehoon Chung wrote:
> > On 11/24/21 4:34 AM, Ulf Hansson wrote:
> > > On Tue, 16 Nov 2021 at 20:02, John Keeping <john@...anate.com> wrote:
> > >>
> > >> The standard stem for MMC aliases is "mmc" and this is used by the MMC
> > >> core to set the slot index.
> > >
> > > This isn't the correct description of the mmc aliases. The below text
> > > is copied from the DT doc:
> > >
> > > "It is possible to assign a fixed index mmcN to an MMC host controller
> > > (and the corresponding mmcblkN devices) by defining an alias in the
> > > /aliases device tree node."
> > >
> > >>
> > >> Use this in preference to the non-standard "mshc" stem when setting the
> > >> controller ID to avoid needing two aliases for each MMC device in order
> > >> to cover both the core and dw_mmc-specific functionality.
> > >>
> > >> The old "mshc" lookup is kept for backwards compatibility.
> > >
> > > The mshc alias is really weird!
> > >
> > > It looks like some leftover from when the dw_mmc controller supported
> > > multiple slots. This support was dropped a long time ago, simply
> > > because it never really worked - and it was not worth trying to. Only
> > > one slot per controller is supported.
> >
> > As Ulf mentioned, dw_mmc controller can be supported multiple slot.
> > But I didn't see its case to use multiple slot. And I had been done to drop a long time ago.
> >
> > mshc was used because of Mobile Storage Host Controller.
> >
> > >
> > > Rather than re-using the mmc alias in the same weird way as the mshc
> > > alias, I suggest we try to remove parsing of the mshc aliases
> > > completely. By looking at the corresponding code and in combination
> > > with the DTS files, it certainly looks doable to me. Do you want to
> > > have a look at it?
> >
> > If possible to remove mshc, it's best.
> > I will check that removing mshc parsing in dw_mmc.c.
>
> Unfortunately it doesn't look like it's easy to remove as there is some
> behaviour depending on this via dw_mci_drv_data::caps, as well as
> different timing setup in dw_mmc-k3.c which uses
> dw_mci_of_alias_get_id() to identify SD and SDIO hosts.
>
> Looking across the dw_mmc-*.c files that use dw_mci_drv_data::caps to
> set capabilities per host controller:
>
> - dw_mmc-exynos.c sets additional capabilities for mshc0, although both
>   MMC_CAP_1_8V_DDR and MMC_CAP_8_BIT_DATA should be set via DT (in fact
>   in some cases it looks like device trees are setting bus-width = <4>
>   so MMC_CAP_8_BIT_DATA seems wrong!); I can't see any device trees
>   setting mmc-ddr-1_8v for these devices at the moment though, so
>   removing that is a change in behaviour

Of course we need to be careful to not break anything. But having some
clever fallback methods could work.

For example, for an SD/SDIO card slot it shouldn't matter if
MMC_CAP_8_BIT_DATA gets set. But I didn't look that closely if that
could be an option.

>
> - dw_mmc-k3.c sets different capabilities for mshc2 and, as mentioned
>   above, uses the alias index to select timing parameters and change
>   voltage switching behaviour

Yeah, the timing thingy looks harder to get rid of for the k3 variant.

Although, if we could limit the alias parsing to the k3 variant, that
would still be a nice cleanup and improvement.

As a next step we could look at introducing some new DT properties for
k3, to specify the timing thingy - and then make the mshc alias
deprecated.

>
> - dw_mmc-hi3798cv200.c and dw_mmc-rockchip.c set the same caps for all
>   slots, so can easily remove the dependency on the alias

Great.

>
>
> I'm mostly interested in Rockchip myself, which is one of the easy ones,
> so I'm not that familiar with Exynos or K3 - I'd guess the Exynos
> version can remove its dependency on the mshc alias pretty easily, but
> the use in dw_mmc-k3.c looks much more difficult given that I can't see
> any other way to derive the necessary info from the current device
> trees.

Right. I will try to get some time to help clean this up.

>
>
> Regards,
> John

Kind regards
Uffe