[RFC] mtd: plat_nand: add further DT bindings and documentation
Brian Norris
computersforpeace at gmail.com
Thu Sep 19 03:28:43 EDT 2013
+ devicetree at vger.kernel.org, remove old list
Hi Alexander,
Sorry for some delay on this. I haven't really had time to look closely
at this previously, and apparently, any users of plat_nand.c are pretty
quiet (or gone?). Your code isn't too invasive, and it seems pretty
reasonable that this simple driver can support platform devices
initialized both via traditional platform structures and via device
tree.
On Sat, Aug 24, 2013 at 07:14:59PM +0100, Alexander Clouter wrote:
> I have run this via the linux-mtd folks before[1], but I get the
> impression everyone is either too busy or does not want to make the
> judgement call about plat_nand's future; linux-mtd pointed me here. :)
>
> I have been busy porting an ARM board (that uses plat_nand) to
> devicetree and stumbled on John Crispin's patch from last year[2].
> There was not much online about how to use it, so I went about
> building upon his work it what seemed to be natural and a good fit.
>
> Attached is the work I have done to the driver its-self (including the
> much needed bindings documentation), whilst on my github page I have
> put up an example of its usage:
>
> https://github.com/jimdigriz/ts78xx/blob/plat-nand/arch/arm/mach-orion5x/board-ts7800.c
> https://github.com/jimdigriz/ts78xx/blob/plat-nand/arch/arm/boot/dts/orion5x-ts7800.dts
>
> Am I going in the right direction here? plat_nand, by its nature, is
> not very DT, the alternative would be a custom driver for my
> platform[3] but then are we not just littering drivers/mtd/nand rather
> than arch/arm/mach-foobar? Is plat_nand going to be marked deprecated?
>
> Meanwhile, the plat_nand driver does currently have some devicetree
> support but it is currently unusable, and more so as there is no
> documentation on how to use it.
>
> The patch below adds the much needed documentation as well as adds some
> common properties (bank-width, chip-delay, bbt-use-flash and nr-chips)
> removing further use of the platform_nand_data struct.
>
> So, what am I to do. Forget plat_nand as it is going to be marked
> deprecated and write a dedicated driver[4]? If not, what do I need to
> do/change/improve to get this ready for for prime time action?
>
> Cheers
>
> [1] http://lists.infradead.org/pipermail/linux-mtd/2013-March/046371.html
> [2] http://lists.infradead.org/pipermail/linux-mtd/2012-April/041025.html
> [3] I could also support the NAND used by arch/arm/mach-ep93xx/ts72xx.c
> but the whole SoC has not been ported to DT
> [4] which ironically it was recommended[5] I used plat_nand when I did this
> (although a lot changes in four years) originally by Hartley who had
> just moved the ts72xx to plat_nand[6]
> [5] http://lists.infradead.org/pipermail/linux-mtd/2009-October/027563.html
> [6] http://lists.infradead.org/pipermail/linux-mtd/2009-February/024555.html
>
> Signed-off-by: Alexander Clouter <alex at digriz.org.uk>
> ---
> .../devicetree/bindings/mtd/plat-nand.txt | 93 ++++++++++++++++++++
> drivers/mtd/nand/plat_nand.c | 53 ++++++++---
> 2 files changed, 134 insertions(+), 12 deletions(-)
> create mode 100644 Documentation/devicetree/bindings/mtd/plat-nand.txt
>
> diff --git a/Documentation/devicetree/bindings/mtd/plat-nand.txt b/Documentation/devicetree/bindings/mtd/plat-nand.txt
> new file mode 100644
> index 0000000..d534bfa
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/mtd/plat-nand.txt
> @@ -0,0 +1,93 @@
> +NAND support for Generic NAND driver
> +
> +Required properties:
> +- compatible : "gen_nand"
> +- reg : Array of base physical addresses of the NAND and the length of memory
> + mapped regions. You *should* have a "nand_data" reg which is the
> + data io region (if not named, then the first reg is used for
> + IO_ADDR_R/IO_ADDR_W), additional regs are platform defined
How might you envision additonal "platform-defined" resources being
included? Are you suggesting they could be utilized by platform code's
bus notifier, as show in your example below?
Also, I wouldn't use the word "should"; either you "must" name the
resource or you "may" (and it's optional). It may also be reasonable to
drop the names altogether and just state that the first reg range must
represent the nand_data.
> +- nr-chips : Number of physical chips
Prefix with "nand," like the others? It's possible this would be a
useful binding for other drivers at some point.
> +
> +Optional properties:
> +- reg-name : "nand_data" *should* be defined, additional ones are platform
> + defined
Again, I don't think you should use "should" here.
> +- bank-width : Width in bytes of the device. Default is 1 (8bit), 2 (16bit)
> + implies NAND_BUSWIDTH_16 and any other value is invalid
Use nand-bus-width. See:
Documentation/devicetree/bindings/mtd/nand.txt
Then you can also re-use the helpers from include/linux/of_mtd.h and
reuse the documentation (just note the supported properties here and
then refer to nand.txt, similar to how you refer to partition.txt).
> +- chip-delay : Chip dependent delay for transferring data from array to read
> + registers in usecs
I guess you might as well prefix this with "nand," as well.
> +- bbt-use-flash : Use a flash based bad block table. Default, OOB identifier
> + is saved in OOB area
There is an existing nand-on-flash-bbt binding in nand.txt for this
purpose.
> +
> +The device tree may optionally contain sub-nodes describing partitions of the
> +address space. See partition.txt for more detail.
> +
> +Example:
> +
> +nand at 800 {
> + #address-cells = <1>;
> + #size-cells = <1>;
> + compatible = "ts,nand", "gen_nand";
> + reg = <0x800 0x04>,
> + <0x804 0x04>;
> + reg-names = "nand_ctrl", "nand_data";
> + nr-chips = <1>;
> + chip-delay = <15>;
> + bbt-use-flash;
> +
> + partition at 0 {
> + label = "mbr";
> + reg = <0x00000000 0x20000>;
> + read-only;
> + };
> +
> + partition at 20000 {
> + label = "kernel";
> + reg = <0x00020000 0x400000>;
> + };
> +
> + partition at 420000 {
> + label = "initrd";
> + reg = <0x00420000 0x400000>;
> + };
> +
> + partition at 820000 {
> + label = "rootfs";
> + reg = <0x00820000 0x1f7e0000>;
> + };
> +};
> +
> +N.B. to use the plat-nand driver, the platform board code does still need to
> + setup platform_nand_data and hook it into the platform_device so
> + the callbacks (for at least .cmd_ctrl and .dev_ready) are available.
> +
> +Example:
> +
> +static struct platform_nand_data ts7800_nand_data = {
> + .ctrl = {
> + .cmd_ctrl = ts7800_nand_cmd_ctrl,
> + .dev_ready = ts7800_nand_dev_ready,
> + },
> +};
> +
> +static int ts7800_platform_notifier(struct notifier_block *nb,
> + unsigned long event, void *__dev)
> +{
> + struct device *dev = __dev;
> +
> + if (event != BUS_NOTIFY_ADD_DEVICE)
> + return NOTIFY_DONE;
> +
> + if (of_device_is_compatible(dev->of_node, "ts,nand"))
> + dev->platform_data = &ts7800_nand_data;
> +
> + return NOTIFY_OK;
> +}
> +
> +static struct notifier_block ts7800_platform_nb = {
> + .notifier_call = ts7800_platform_notifier,
> +};
> +
> +void __init ts7800_init(void)
> +{
> + bus_register_notifier(&platform_bus_type, &ts7800_platform_nb);
> +}
I'm not quite sure what to say about this platform code example. It may
be an acceptable route if that's the only option. But there may be
alternatives that don't involve relegating this to platform board code.
Perhaps the platform_data callback routines could be provided for
device-tree enabled systems via the plat_nand driver's of_match_table
(matching against, e.g., "ts,nand")? Basically, the of_device_id.data
subfield could hold a platform_nand_ctrl struct.
> diff --git a/drivers/mtd/nand/plat_nand.c b/drivers/mtd/nand/plat_nand.c
> index c004566..1407db5 100644
> --- a/drivers/mtd/nand/plat_nand.c
> +++ b/drivers/mtd/nand/plat_nand.c
> @@ -12,6 +12,7 @@
> #include <linux/io.h>
> #include <linux/module.h>
> #include <linux/platform_device.h>
> +#include <linux/of.h>
> #include <linux/slab.h>
> #include <linux/mtd/mtd.h>
> #include <linux/mtd/nand.h>
> @@ -23,7 +24,7 @@ struct plat_nand_data {
> void __iomem *io_base;
> };
> -static const char *part_probe_types[] = { "cmdlinepart", NULL };
> +static const char *part_probe_types[] = { "cmdlinepart", "ofpart", NULL };
Since these are the defaults (see mtdpart.c, default_mtd_part_types),
you can just remove this and pass NULL instead.
> /*
> * Probe for the NAND device.
> @@ -42,14 +43,11 @@ static int plat_nand_probe(struct platform_device *pdev)
> return -EINVAL;
> }
> - if (pdata->chip.nr_chips < 1) {
> - dev_err(&pdev->dev, "invalid number of chips specified\n");
> - return -EINVAL;
> - }
> -
> - res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + res = platform_get_resource_byname(pdev, IORESOURCE_MEM, "nand_data");
> if (!res)
> - return -ENXIO;
> + res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> + if (!res)
> + return -ENXIO;
This fall-back (from get_resource_byname() to get_resource()) is a
little awkward IMO. This relates to the discussion above--are we going
to require a resource name? If not, then I think we don't even want to
get the resource by name at all; even if it has an (optional) name, we
just get it from index 0.
> /* Allocate memory for the device structure (and zero it) */
> data = kzalloc(sizeof(struct plat_nand_data), GFP_KERNEL);
> @@ -79,15 +77,40 @@ static int plat_nand_probe(struct platform_device *pdev)
> data->chip.IO_ADDR_R = data->io_base;
> data->chip.IO_ADDR_W = data->io_base;
> +
> + if (pdev->dev.of_node) {
> + int i;
> +
> + if (!of_property_read_u32(pdev->dev.of_node,
> + "nr-chips", &i))
> + data->chip.numchips = i;
> + if (!of_property_read_u32(pdev->dev.of_node,
> + "chip-delay", &i))
> + data->chip.chip_delay = (u8)i;
> + if (!of_property_read_u32(pdev->dev.of_node,
> + "bank-width", &i)) {
of_get_nand_bus_width()?
> + if (i == 2)
> + data->chip.options |= NAND_BUSWIDTH_16;
> + else if (i != 1) {
> + dev_warn(&pdev->dev,
> + "%d bit bus width out of range\n", i);
> + }
> + }
> + if (of_get_property(pdev->dev.of_node, "bbt-use-flash", &i))
of_get_nand_on_flash_bbt()?
> + data->chip.bbt_options |= NAND_BBT_USE_FLASH;
> + } else {
> + data->chip.numchips = pdata->chip.nr_chips;
> + data->chip.chip_delay = pdata->chip.chip_delay;
> + data->chip.options |= pdata->chip.options;
> + data->chip.bbt_options |= pdata->chip.bbt_options;
> + }
> +
> data->chip.cmd_ctrl = pdata->ctrl.cmd_ctrl;
> data->chip.dev_ready = pdata->ctrl.dev_ready;
> data->chip.select_chip = pdata->ctrl.select_chip;
> data->chip.write_buf = pdata->ctrl.write_buf;
> data->chip.read_buf = pdata->ctrl.read_buf;
> data->chip.read_byte = pdata->ctrl.read_byte;
> - data->chip.chip_delay = pdata->chip.chip_delay;
> - data->chip.options |= pdata->chip.options;
> - data->chip.bbt_options |= pdata->chip.bbt_options;
> data->chip.ecc.hwctl = pdata->ctrl.hwcontrol;
> data->chip.ecc.layout = pdata->chip.ecclayout;
> @@ -102,8 +125,14 @@ static int plat_nand_probe(struct platform_device *pdev)
> goto out;
> }
> + if (data->chip.numchips < 1) {
> + dev_err(&pdev->dev, "invalid number of chips specified\n");
> + err = -EINVAL;
> + goto out;
> + }
> +
> /* Scan to find existence of the device */
> - if (nand_scan(&data->mtd, pdata->chip.nr_chips)) {
> + if (nand_scan(&data->mtd, data->chip.numchips)) {
> err = -ENXIO;
> goto out;
> }
Brian
More information about the linux-mtd
mailing list