From mboxrd@z Thu Jan  1 00:00:00 1970
From: linux@arm.linux.org.uk (Russell King - ARM Linux)
Date: Thu, 3 Oct 2013 20:36:23 +0100
Subject: IMX6 USDHC card detection
Message-ID: <20131003193623.GF12758@n2100.arm.linux.org.uk>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Having spent many hours chasing around wondering why controller based
card detection doesn't work on IMX6, I've finally found it.

The hardware is setup such that we have a card detect signal routed to
GPIO 4.  If we try and use the host controller's built-in card detection,
things go awry - the host controller believes that a card is always
inserted.

This appears to be because the kernel sets the D3CD bit in the PROT_CTRL
register.  This enables card detection via the DAT3 signal, which
indicates a logic '1' level when a card is inserted.  SD cards have a 50k
pullup internally which provides the card detection mechanism, and hosts
are expected to have a weak pull-down on this signal.

However, the default setup of the SD data lines in IOMUX is this (eg):

             MX6QDL_PAD_SD2_DAT0__SD2_DATA0 0x17059
             MX6QDL_PAD_SD2_DAT1__SD2_DATA1 0x17059
             MX6QDL_PAD_SD2_DAT2__SD2_DATA2 0x17059
             MX6QDL_PAD_SD2_DAT3__SD2_DATA3 0x17059

Bits 14 and 15 indicate "pull this pin up via a 22K resistor".

What you end up with is the SD controller believes that a card is always
inserted no matter what you do.  You can see this through the PRES_STATE
register - you see bit 18 assert/deassert appropriately on card insertion
and removals indicating that the socket CD signal is changing state, but
you don't see the card inserted bit (bit 16) ever change.  Neither does
the kernel react in any way.

So, are the defaults for DATA3 signals really appropriate, or should they
be adjusted (in other words, 0x13059 for DATA3).  The alternative is to
ignore the built-in host card detection, and configure the CD pin as a
GPIO instead.