Re: [PATCH v13 07/12] drm: bridge/dw_hdmi: add support for multi-byte register width access

From: Philipp Zabel <p.zabel@pengutronix.de>
To: Andy Yan <andy.yan@rock-chips.com>
Cc: Mark Rutland <mark.rutland@arm.com>,
	heiko@sntech.de, dri-devel@lists.freedesktop.org,
	ykk@rock-chips.com, devel@driverdev.osuosl.org,
	Pawel Moll <pawel.moll@arm.com>,
	linux-rockchip@lists.infradead.org,
	Grant Likely <grant.likely@linaro.org>,
	Dave Airlie <airlied@redhat.com>,
	jay.xu@rock-chips.com, devicetree@vger.kernel.org,
	Zubair.Kakakhel@imgtec.com, Arnd Bergmann <arnd@arndb.de>,
	Ian Campbell <ijc+devicetree@hellion.org.uk>,
	Rob Herring <robh+dt@kernel.org>,
	rmk+kernel@arm.linux.org.uk, mark.yao@rock-chips.com,
	fabio.estevam@freescale.com, Josh Boyer <jwboyer@redhat.com>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	linux-kernel@vger.kernel.org, djkurtz@google.com,
	Kumar Gala <galak@codeaurora.org>
Subject: Re: [PATCH v13 07/12] drm: bridge/dw_hdmi: add support for multi-byte register width access
Date: Mon, 01 Dec 2014 13:04:16 +0100	[thread overview]
Message-ID: <1417435456.4624.12.camel@pengutronix.de> (raw)
In-Reply-To: <547843B9.6000806@rock-chips.com>

Am Freitag, den 28.11.2014, 17:43 +0800 schrieb Andy Yan:
> Hi Zabel:
> On 2014年11月27日 00:34, Philipp Zabel wrote:
> > Am Mittwoch, den 26.11.2014, 21:32 +0800 schrieb Andy Yan:
> >> On rockchip rk3288, only word(32-bit) accesses are
> >> permitted for hdmi registers.  Byte width accesses (writeb,
> >> readb) generate an imprecise external abort.
> >>
> >> Signed-off-by: Andy Yan <andy.yan@rock-chips.com>
> >>
> >> ---
> >>
> >> Changes in v13: None
> >> Changes in v12: None
> >> Changes in v11: None
> >> Changes in v10: None
> >> Changes in v9: None
> >> Changes in v8: None
> >> Changes in v7: None
> >> Changes in v6:
> >> - refactor register access without reg_shift
> >>
> >> Changes in v5:
> >> - refactor reg-io-width
> >>
> >> Changes in v4: None
> >> Changes in v3:
> >> - split multi-register access to one indepent patch
> >>
> >>   drivers/gpu/drm/bridge/dw_hdmi.c | 57 +++++++++++++++++++++++++++++++++++-----
> >>   1 file changed, 51 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/drivers/gpu/drm/bridge/dw_hdmi.c b/drivers/gpu/drm/bridge/dw_hdmi.c
> >> index a53bf63..5e88c8d 100644
> >> --- a/drivers/gpu/drm/bridge/dw_hdmi.c
> >> +++ b/drivers/gpu/drm/bridge/dw_hdmi.c
> >> @@ -100,6 +100,11 @@ struct hdmi_data_info {
> >>   	struct hdmi_vmode video_mode;
> >>   };
> >>   
> >> +union dw_reg_ptr {
> >> +	u32 __iomem *p32;
> >> +	u8 __iomem *p8;
> >> +};
> > I see no need to introduce this. Just explicitly multiply the offset in
> > dw_hdmi_writel.
> >
>      Is there any disadvantage to do like this?
>      The compiler can help us do the explicitly multiply by this way.

Four additional lines, a new defined type, a few more changes to struct
dw_hdmi and dw_hdmi_bind necessary.

Technically I see no problem to let the compiler do the multiplication,
my issue is that it ever so slightly obfuscates the code. Instead of
just writing "* 4" in two functions, we get a new union that you need to
know about when looking at struct dw_hdmi and dw_hdmi_bind, regs.p8 is
used but never assigned directly, it's just a tiny bit of additional
effort needed to understand the code. But when the cost to avoid that is
so small...

regards
Philipp

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel