RE: [PATCH 04/14] mmc: renesas_sdhi: Add support for R-Car H3e-2G and M3e-2G

From: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
To: Wolfram Sang <wsa@kernel.org>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>,
	Laurent Pinchart <laurent.pinchart@ideasonboard.com>,
	Magnus Damm <magnus.damm@gmail.com>,
	Rob Herring <robh+dt@kernel.org>,
	Linux-Renesas <linux-renesas-soc@vger.kernel.org>,
	"open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" 
	<devicetree@vger.kernel.org>
Subject: RE: [PATCH 04/14] mmc: renesas_sdhi: Add support for R-Car H3e-2G and M3e-2G
Date: Thu, 24 Jun 2021 10:54:34 +0000	[thread overview]
Message-ID: <TY2PR01MB36927B0CCE7C557A3115E481D8079@TY2PR01MB3692.jpnprd01.prod.outlook.com> (raw)
In-Reply-To: <YNQi0w4zsG01ezgu@kunai>

Hi Wolfram-san,

> From: Wolfram Sang, Sent: Thursday, June 24, 2021 3:15 PM
> 
> Hi all,
> 
> > > > > --- a/drivers/mmc/host/renesas_sdhi_core.c
> > > > > +++ b/drivers/mmc/host/renesas_sdhi_core.c
> > > > > @@ -943,6 +943,8 @@ static const struct soc_device_attribute sdhi_quirks_match[]  = {
> > > > >       { .soc_id = "r8a77965", .data = &sdhi_quirks_r8a77965 },
> > > > >       { .soc_id = "r8a77980", .data = &sdhi_quirks_nohs400 },
> > > > >       { .soc_id = "r8a77990", .data = &sdhi_quirks_r8a77990 },
> > > > > +     { .soc_id = "r8a779m1", .data = &sdhi_quirks_bad_taps2367 },
> > > > > +     { .soc_id = "r8a779m3", .data = &sdhi_quirks_bad_taps1357 },
> > > >
> > > > Could we reuse the entries for H3 and M3 instead, by dropping the
> > > > "ES3.*" revision ?
> > >
> > > We cannot reuse the H3 ES3.0 entry, as soc_device_match()
> > > works differently than of_machine_is_compatible(): the former doesn't
> > > consider "r8a779m1" and "r8a7795" equivalent, the latter does.
> > > Same for M3-W+ (no explicit ES3.0 there) and M3e-2G.
> > >
> > > It's a pity we still don't have a "quirk-free" SDHI version on H3
> > > and M3-W class SoCs (waiting for ES4.0?), as that would allow us to
> > > just match on "renesas,sdhi-r8a7795" resp. "renesas,sdhi-r8a77961"
> > > through the driver's .of_match_table[] instead, which would work for
> > > H3e-2G and M3e-2G, too.
> >
> > Perhaps, ES4.0 will not be released. So, we can refactor the driver's
> > .of_match_table[] now. I investigated this a little, and it seems
> > we need many renesas_sdhi_of_data for each SoC instead of
> > of_rcar_gen3_compatible. But, I guess such modification is better
> > than adding sdhi_quirks_match entries.
> >
> > Wolfram-san, what do you thinks?
> 
> I don't fully understand how the refactoring should look like? Is it
> moving 'struct renesas_sdhi_quirks' to renesas_sdhi_internal_dmac.c and
> merge it there with renesas_sdhi_of_data? Is it really better to copy
> this struct per SoC? Most of the data is the same.

I also have the same concern. But, I guess we can refactor
the renesas_sdhi_of_data like below to avoid increasing data size:

struct renesas_sdhi_of_data_with_quirks {
	const struct renesas_sdhi_of_data *of_data;
	const struct renesas_sdhi_quirks *quirks;
};

And then, we can keep of_rcar_gen3_compatible and
we can add each SoC's renesas_sdhi_of_data_with_quirks
and set it to the .data.

Best regards,
Yoshihiro Shimoda

> Thanks,
> 
>    Wolfram