On 08:14 Sat 30 Apr     , Ryan Mallon wrote:

On 30/04/11 06:04, H Hartley Sweeten wrote:

quoted

On Thursday, April 28, 2011 7:59 PM, Ryan Mallon wrote:

quoted

Each AT91 variant (AT91RM9200, AT91SAM9260, etc) currently has its own
devices file, which includes the MMIO address, interrupt
configuration, GPIO setup, etc for each device. This results in a large
amount of duplicated code.

This patch set introduces a framework for adding shared devices for the
AT91 platform and replaces the multiple device setup implementations for
each device with single implementations in the new framework. This has
a net reduction of nearly 5000 lines of code.

Each of the arch/arm/mach-at91/*_devices.c files becomes a collection of
structures (with some initialisation callbacks where necessary) with a
table of devices which are present on the particular AT91 variant. All
structures/functions are marked as __init/__initdata so there is little
additional memory overhead. This also means that the #ifdefs around each
device can be removed from the *_devices.c files (but remain in the new
common devices.c file) without overhead.

This patch series does not introduce any functional changes to how the
board files add devices, it only replaces the duplicate device 
initialisation code with common versions. The patch series attempts to
have minimum change by rewriting as little as possible of the actual
device initialisation functions.

This is also a step towards allowing more than one AT91 variant to be
built into a single kernel by removing duplicate function names across
the *_devices.c files.

I have build tested the patch series for all of the  AT91 variants 
and devices, and have boot tested it on the AT91SAM9G20 (Snapper 9G20
board) and tested basic device functionality.

Ryan,

This patch series does not apply to linux-next.

It appears Jean-Christophe PLAGNIOL-VILLARD has recently committed a number
of patches to that branch.

    at91: move clock subsystem init to soc generic init
    at91: move register clocks to soc generic init
    at91: move st timer to drivers/clocksource
    at91: move pit timer to drivers/clocksource
    at91: switch st timer to early platform devices
    at91: switch pit timer to early platform devices
    at91: move gpio to drivers/gpio
    at91: switch gpio to early platfrom device
    at91: switch to CLKDEV_LOOKUP
    at91: use structure to store the current soc
    at91: merge board usb-a9260 and usb-a9263 together
    at91: factorize at91 interrupts init to soc
    at91: introduce commom AT91_BASE_SYS
    at91rm9200: introduce at91rm9200_set_type to specficy cpu package
    at91: 9260 and 9g20 add support of join SRAM Memory Mapping
    at91/board-eco920: remove at91_beeper ressource as no driver at91_beeper exist
    atmel_serial: keep the platform_device unchanged
    at91: remove MTD_NAND_ATMEL_BUSWIDTH_16 option
    at91: Add ARCH_ID and basic cpu macros definition for 5series chips family.

There are also these three ahead of Linus' linux-2.6 tree:

    arm: at91: fix compiler warning for eb01 board build
    ARM: at91: AT91CAP9 has a macb device
    treewide: remove extra semicolons

Ok, I have more changes to make anyway so I will rebase against these
three. I do not want to rebase on top of the Jean-Christophe's patches
because I do not think that they are ready for mainline inclusion. See
the discussion in the other thread.

The soc fix need to first, the devices resource after
as the alot of devices resources need to be fix and api changes before such as
RTC, GPIO, AT91_BASE_SYS, switch to clkdev, board init etc...

otherwise the factorisation of the resources will not be able to be tested and
we need to re-updated due to design issue.

So please wait we finish this fix first as they are bigger issues

And this will allow to test the factorisation more easly and ensure we break
nothing

Best Regards,
J.