Re: [PATCH 1/2] sdhci-of-arasan: Add quirk and device tree parameter to fake CD bit

From: Zach Brown <zach.brown@ni.com>
To: Mark Rutland <mark.rutland@arm.com>
Cc: <adrian.hunter@intel.com>, <robh+dt@kernel.org>,
	<ulf.hansson@linaro.org>, <linux-mmc@vger.kernel.org>,
	<devicetree@vger.kernel.org>, <linux-kernel@vger.kernel.org>,
	<michal.simek@xilinx.com>, <soren.brinkmann@xilinx.com>,
	<linux-arm-kernel@lists.infradead.org>, <lars@metafoo.de>
Subject: Re: [PATCH 1/2] sdhci-of-arasan: Add quirk and device tree parameter to fake CD bit
Date: Thu, 25 Aug 2016 13:26:22 -0500	[thread overview]
Message-ID: <20160825182621.GA14817@zach-desktop> (raw)
In-Reply-To: <20160825180959.GA14705@remoulade>

On Thu, Aug 25, 2016 at 07:10:00PM +0100, Mark Rutland wrote:
> On Thu, Aug 25, 2016 at 12:15:44PM -0500, Zach Brown wrote:
> > On Thu, Aug 25, 2016 at 11:56:55AM +0100, Mark Rutland wrote:
> > > On Wed, Aug 24, 2016 at 06:23:03PM -0500, Zach Brown wrote:
> > > > +- fake-cd: On Zynq Devices the SDHCI Controller will not work without the cd
> > > > +  bit. When this option is set the driver will put the controller in test mode
> > > > +  and fake the cd bit so it will function.
> > > 
> > > As Lars noted, the DT should describe the HW, and the policy of how to deal
> > > with that should be left to the kernel. So from a DT perspective the above is
> > > not correct.
> > > 
> > > If I understand the linked documentation, this is slightly different to typical
> > > uses of broken-cd in that in the absence of a card detect signal the HW will
> > > not be able to access the SD card at all, even if requested to. Is that correct?
> > > 
> > > If so, perhaps a better option is to have the combination of broken-cd and the
> > > compatible string for this IP block imply that the test mode workaround is
> > > required. Obviously that requires a fixup to the usual broken-cd binding to
> > > remove the implication that polling alone must be used.
> > > 
> > > Thanks,
> > > Mark.
> > 
> > In cases where the card is non-removable then polling doesn't make sense.
> 
> We have the non-removable property to describe that, so we can also look at that.
> 
> > So it doesn't make sense to tie the test mode workaround into the broken-cd
> > property, even though I agree the nature of the defect fits under the notion
> > of the CD being broken.
> 
> Maybe not solely on broken-cd, but I think that we dont necessarily need a new
> DT property. As above, broken-cd, non-removable, and the compatible string may
> together give the kernel enough information to choose the right thing to do.
> 
> Thanks,
> Mark.

I'm not sure if I understand your suggestion completely. Are you suggesting
setting both the broken-cd and non-removable properties? That would make sense,
but my understanding was that the two properities are not meant to co-exist. In
/Documentation/devicetree/bindings/mmc/mmc.txt it states that only one should
be supplied. Don't the two properties conflict with each other?