Speed up transfers between NAND and RAM by using the io{read,write}*_rep()
calls to replace a more limited read-only I/O loop.  On a similar arm926
board this gave a significant speedup (ISTR 16%) in a userspace "dd"
benchmark.

Also add a comment explaining some key assumptions (wiring) this makes.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

That define was only used for polling for card presence,
which now we let mmc core do it in a more reliable way.

No need to keep that define here.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

There's no need for a file included only once, remove it
and move the register definition to davinci_mmc.c.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Make rw_threshold and use_dma a module parameter,
thus getting rid of struct mmcsd_config_def.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Add missing __init and __exit where appropriate.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Mostly tab fixes, but also removing some extra lines and
putting module_init() close to its argument function.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

On Mon, Dec 15, 2008 at 09:19:05PM +0200, Felipe Balbi wrote:
> cool, will fix :-)
>
> good catch with the missing branch, thanks

here you are

==================== cut here ====================

>From 43fbab69f7d19e39c7c270c4aa1652b0efea0a8f Mon Sep 17 00:00:00 2001
From: Felipe Balbi <felipe.balbi@nokia.com>
Date: Mon, 15 Dec 2008 18:08:36 +0200
Subject: [patch-v2.6.28 3/8] mmc: host: davinci: reimplement read/write fifo in C

It's easier to follow and to maintain.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

There were a bunch of useless function prototypes in
davinci_mmc.h. Remove them and reorganize the file
so we don't really need those.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Fix most of checkpatch.pl ERRORS with davinci_mmc.c. Also
fix up a few comments in the beggining of the file.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Fix checkpatch.pl bugs ... mostly return-with-parens.  And also
fix some whitespace bugs, one of which was reported by checkpatch
(bad indentation).
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Update more of the RX and TX buffer descriptor (BD) handling to
properly distinguish between __iomem pointers and normal data
memory address space, getting rid of 32 "cast removes address
space of expression" messages.

This patch doesn't resolve all the issues sparse reports.  In fact
there are now 48 new "dereference of noderef expression" messages.
They mean that a patch is needed to convert BD member accessors to
use readl/writel calls.

Plus, there are several related issues that sparse won't warn about.
The initial i/o space addresses -- SRAM used for BDs, and controller
registers -- should come from ioremap.  Reverse mappings shouldn't
usually be needed, and if they are they should avoid io_v2p() and
virt_to_phys() calls.  And the need to use casts so often is wrong,
especially with __force annotations.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Fix about half the warnings from "make C=1" (sparse), and start on
the __iomem issues associated with the CPPI buffer descriptor data.
All of the remaining warnings come from address space issues.  For
this driver, fixing all those issues will be quite messy.

One of these warnings highlighted minor goofiness in how the module
infrastructure was used; fix that too.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Do spinlock initialization during davinci_emac::probe(), when
the lock is allocated, not each time it's opened.  That's the
correct place to do such work.

More significantly, fix a bug: one of the three spinlocks was
never getting initialized, preventing the system from booting
in some cases (e.g. running with lockdep).
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

CC      drivers/net/davinci_emac.o
drivers/net/davinci_emac.c: In function 'emac_adjust_link':
drivers/net/davinci_emac.c:2323: error: incompatible type for argument 1 of '_spin_lock_irqsave'
make[2]: *** [drivers/net/davinci_emac.o] Error 1
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

NAND devices can't be hotpluggable. We can use that
to shrink code size a bit by using platform_driver_probe().
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

Those mux settings are only valid for dm6446, so put that
under a conditional.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

Cleanup only. Removing whitespaces from <mach/nand.h>
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

Make dm355's nand flash probe with current driver.
For some reason, the current driver marks way too many
blocks as bad blocks. Later patches will be needed to
fix it.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

On Sat, Dec 13, 2008 at 12:54:17AM +0200, Felipe Balbi wrote:
> On Fri, Dec 12, 2008 at 02:52:14PM -0800, David Brownell wrote:
> > Just update the individual patches and save the effort.
> > Less work for you, fewer bits wasted on the Internet.  ;)
>
> Sure, just did it. Will retest (sanity) and reply with the newer patch
> soon.

here it is:

==== cut here ====

>From 964d8b463ea8d32c27249d6e1b64ed6143551dd8 Mon Sep 17 00:00:00 2001
From: Felipe Balbi <felipe.balbi@nokia.com>
Date: Wed, 10 Dec 2008 01:41:52 +0200
Subject: [patch-v2.6.28 06/10] mtd: nand: davinci: add missing __init and __exit

Add __init to init-only functions, __exit to exit-only
functions.

Also, switch two __devinit to __init, as nand devices
can't be hotpluggable.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

add a more meaningful set of debugging messages.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

On Fri, Dec 12, 2008 at 07:52:28PM +0200, Felipe Balbi wrote:
> From: Felipe Balbi <felipe.balbi@nokia.com>
>
> introduce a structure to hold davinci_nand device
> data.
>
> Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

New version below:

====== cut here ======

>From 1d2f9999a56bd0a71510ce09e0e863d15d98013d Mon Sep 17 00:00:00 2001
From: Felipe Balbi <felipe.balbi@nokia.com>
Date: Fri, 12 Dec 2008 16:15:07 +0200
Subject: [patch-v2.6.28 04/10] mtd: nand: introduce davinci_nand_info

introduce a structure to hold davinci_nand device
data.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

nand_vaddr was already ioremaped, so it's already
a void __iomem pointer. Do not cast it uselessly.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

checkpatch.pl fixes to davinci_nand.c
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

Later patch will come to use it in davinci_nand.c and get
rid of a define there. In DM355, the base is different, so
better to apply this patch before adding support for DM355
nand chip.
Signed-off-by: Felipe Balbi <felipe.balbi@nokia.com>

Checkpatch likes EXPORT_SYMBOL below function definition.
Also, add export of davinci_pause_dma and davinci_resume_dma.
Signed-off-by: Troy Kisky <troy.kisky@boundarydevices.com>

Don't clear link on stop until audio driver is fixed.
Signed-off-by: Troy Kisky <troy.kisky@boundarydevices.com>

This shrinks the object size of davinci_mmc by about a quarter,
ripping out stuff that was issuing all kinds of commands behind
the back of the MMC core (evidently for card detect purposes).

That's stuff that needs to be ripped out before this driver can go
into mainline, obviously; good to remove it ASAP.

This has one immediate positive effect:  cards which previously could
not be enumerated on a dm355evm can now enumerate!  (After disabling
DMA, for now.)  Which in turn let me boot a rootfs from an SD card ...

There still seems to be state kept around after card removal, which
prevents a second card from enumerating after removing the first one.
Also, the MMC core failed when enumerating an MMC card; odd!

(Plus fix two broken debug messages:  voltages use a bitmask, and
newlines don't belong in the *middle* of messages.)
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>

Conflicts:
	drivers/i2c/chips/tlv320aic23.c

Cleanup the way the memory regions are intialised and removes the hard-coded
offsets used for EMAC internal modules (MDIO, Control RAM, EMAC wrapper)
i.e the assumption that all modules are at fixed offset from the EMAC "base".

Also removes the hard coded values for the RAM size, number of transmit and
receive buffer descriptors, instead gets the info from the driver structure.
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Modifies the EMAC resource data structure by adding an entry to each EMAC
memory region.
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Removes 'davinci_emac_phy.c' and  'davinci_emac_phy.h' files
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Registers a board fixup for Intel LXT971 PHY on DaVinci to increase the transmit
signal strength - a fix for transmit lockup issue.
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Adds  MII/PHY layer support DM644x/DM646x
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Modifies davinci_all_defconfig to support PHY Layer
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Modifies the davinci_evm_dm646x_defconfig to support PHY Layer.
Signed-off-by: Chaithrika U.S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Modifies davinci_evm_dm644x_defconfig to support PHY Layer
Signed-off-by: Chaithrika U S <chaithrika@ti.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Remove a header inclusion that was accidentaly
left over from debugging.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Minor cleanups to davinci-mmc: whitespace, spelling, sparse fixes,
duplicated include files, broken debug message, ioremap exactly
the address space we reserved, cleaner IS_ERR check, etc.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Support card detect and writeprotect switches on DM355 EVM.
Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>

Replace the volatile structure references with
__raw_writel and __raw_readl functions.
Signed-off-by: Troy Kisky <troy.kisky@boundarydevices.com>
Signed-off-by: Kevin Hilman <khilman@deeprootsystems.com>