Re: [PATCH] Revert "mtd: davinci-nand: disable subpage write for keystone-nand"

[Boris Brezillon <boris.brezillon@free-electrons.com>]

On Thu, 30 Mar 2017 13:13:58 +0530
Sekhar Nori <nsekhar@ti.com> wrote:

> Hi Boris,
> 
> On Wednesday 29 March 2017 08:33 PM, Boris Brezillon wrote:
> > On Mon, 27 Mar 2017 15:56:55 +0530
> > Sekhar Nori <nsekhar@ti.com> wrote:
> >   
> >> This partly reverts commit 28c015a9daabe4ed3aeb0ccf669a3f1c2b8b81d5.
> >>
> >> Since f6d7c1b5598b ("mtd: nand: davinci: Reinitialize the HW ECC
> >> engine in 4bit hwctl") using subpage and 4bit HWECC works.
> >>
> >> Revert the original workaround made for this. Tested with
> >> mtd_subpagetest and UBIFS on K2E, K2HK, K2L EVMs.
> >>
> >> The addition of keystone specific compatible to davinci_nand_of_match[]
> >> contained in original commit is retained.  
> > 
> > Hm, it sounds like a bad idea to re-enable sub-page writes afterwards.
> > You might break existing systems using UBI, because UBI detects the
> > sub-page write feature and decides to place the VID header at a
> > different position depending on that.
> > To sum-up, you're likely to break users that have already flashed a UBI
> > image with the old driver.  
> 
> Okay, I didn't realize that.
> 
> > If you want to re-enable sub-page writes, just switch back to the
> > "ti,davinci-nand" compatible.  
> 
> How about a comment on top of the disable subpage write code explaining 
> that its not really needed anymore ? Something like this patch:
> 
> ---8<---
> From: Sekhar Nori <nsekhar@ti.com>
> Date: Thu, 30 Mar 2017 13:08:30 +0530
> Subject: [PATCH] mtd: nand: davinci: add comment on NAND subpage write status
>  on keystone
> 
> Add a comment clarifying that NAND subpage write on keystone works,
> but is not being enabled in the interest of backward compatibility.
> 
> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> ---
>  drivers/mtd/nand/davinci_nand.c | 11 +++++++++++
>  1 file changed, 11 insertions(+)
> 
> diff --git a/drivers/mtd/nand/davinci_nand.c b/drivers/mtd/nand/davinci_nand.c
> index 27fa8b87cd5f..5bf8be1cfc8c 100644
> --- a/drivers/mtd/nand/davinci_nand.c
> +++ b/drivers/mtd/nand/davinci_nand.c
> @@ -581,6 +581,17 @@ static struct davinci_nand_pdata
>  			"ti,davinci-nand-use-bbt"))
>  			pdata->bbt_options = NAND_BBT_USE_FLASH;
>  
> +		/*
> +		 * Since kernel v4.8, this driver has been fixed to enable
> +		 * use of 4-bit hardware ECC with subpages and verified on
> +		 * TI's keystone EVMs (K2L, K2HK and K2E).
> +		 * However, in the interest of not breaking systems using
> +		 * existing UBI partitions, sub-page writes are not being
> +		 * (re)enabled. If you want to use subpage writes on Keystone
> +		 * platforms (i.e. do not have any existing UBI partitions),
> +		 * then remove the if() block below or pass "ti,davinci-nand"
> +		 * as the compatible in your device-tree file.

Hm, I'm not sure we should suggest to remove these lines from the code.
Just suggest to use "ti,davinci-nand" compatible.
If you're okay with that, no need to resend, I can rework the comment
accordingly before applying the patch.

Another long term solution: if UBI is the only MTD user to be
incompatible when sub-page write is activated/deactivated, maybe we can
relax the checks in UBI to allow it to attach a partition where VID
header offset is at mtd->writesize instead of
mtd->writesize >> mtd->subpage_sft.

I'll let Richard comment on that solution though.

> +		 */
>  		if (of_device_is_compatible(pdev->dev.of_node,
>  					    "ti,keystone-nand")) {
>  			pdata->options |= NAND_NO_SUBPAGE_WRITE;