Re: [PATCH 2/3] ARM: i.MX: implement GPMI NAND xload

[Sascha Hauer <sha@pengutronix.de>]

On Tue, Jan 26, 2021 at 12:40:17PM +0100, Andrej Picej wrote:
> 
> On 25. 01. 21 10:43, Andrej Picej wrote:
> > On 25. 01. 21 10:15, Sascha Hauer wrote:
> > > On Thu, Jan 21, 2021 at 11:28:39AM +0100, Andrej Picej wrote:
> > > > Hi Sascha,
> > > > 
> > > > thanks for your comments.
> > > > 
> > > > Responses inline.
> > > > 
> > > > On 21. 01. 21 10:01, Sascha Hauer wrote:
> > > > > Hi Andrej,
> > > > > 
> > > > > Some comments inline.
> > > > > 
> > > > > On Wed, Jan 20, 2021 at 01:51:06PM +0100, Andrej Picej wrote:
> > > > > > From: Sascha Hauer <s.hauer@pengutronix.de>
> > > > > > 
> > > > > > Commit is based on initial Sascha Hauer's work. It
> > > > > > implements PBL xload
> > > > > > mechanism to load image from GPMI NAND flash.
> > > > > > 
> > > > > > Additional work was done, so that the NAND's size, page size and OOB's
> > > > > > size are autodetected and not hardcoded. Detection method follows the
> > > > > > same methods as used in NAND driver, meaning NAND ONFI
> > > > > > support is probed
> > > > > > and if NAND supports ONFI, NAND memory organization is read from ONFI
> > > > > > parameter page otherwise "READ ID" is used.
> > > > > > 
> > > > > > Currently only implemented for i.MX6 familly of SoCs.
> > > > > > 
> > > > > > Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
> > > > > > Signed-off-by: Primoz Fiser <primoz.fiser@norik.com>
> > > > > > Signed-off-by: Andrej Picej <andrej.picej@norik.com>
> > > > > > ---
> > > > > >    arch/arm/mach-imx/Makefile             |    2 +-
> > > > > >    arch/arm/mach-imx/include/mach/xload.h |    1 +
> > > > > >    arch/arm/mach-imx/xload-gpmi-nand.c    | 1163
> > > > > > ++++++++++++++++++++++++
> > > > > >    3 files changed, 1165 insertions(+), 1 deletion(-)
> > > > > >    create mode 100644 arch/arm/mach-imx/xload-gpmi-nand.c
> > > > > > 
> > > > > > diff --git a/arch/arm/mach-imx/Makefile b/arch/arm/mach-imx/Makefile
> > > > > > index e45f758e9..d94c846a1 100644
> > > > > > --- a/arch/arm/mach-imx/Makefile
> > > > > > +++ b/arch/arm/mach-imx/Makefile
> > > > > > +static int mxs_nand_get_info(struct mxs_nand_info
> > > > > > *info, void *databuf)
> > > > > > +{
> > > > > > +    int ret, i;
> > > > > > +
> > > > > > +    ret = mxs_nand_check_onfi(info, databuf);
> > > > > > +    if (ret) {
> > > > > > +        if (ret != 1)
> > > > > > +            return ret;
> > > > > > +        pr_warn("ONFI not supported, try \"READ ID\"...\n");
> > > > > 
> > > > > You already printed a "ONFI not supported\n" message. Printing it once
> > > > > is enough. Also this message appears with every non-ONFI
> > > > > nand, right? In
> > > > > that case it should rather be pr_info()
> > > > 
> > > > OK, I agree, will fix it in v2.
> > > > 
> > > > > 
> > > > > > +    /*
> > > > > > +     * If ONFI is not supported or if it fails try to
> > > > > > get NAND's info from
> > > > > > +     * "READ ID" command.
> > > > > > +     */
> > > > > > +    pr_debug("Trying \"READ ID\" command...\n");
> > > > > > +    ret = mxs_nand_get_readid(info, databuf);
> > > > > > +    if (ret) {
> > > > > > +        if (ret != -EOVERFLOW)
> > > > > > +            return ret;
> > > > > > +
> > > > > > +        /*
> > > > > > +         * If NAND's "READ ID" returns bad values, try to set them to
> > > > > > +         * default (most common NAND memory org.) and continue.
> > > > > > +         */
> > > > > > +        pr_warn("NANDs READ ID command returned bad values" \
> > > > > > +            " set them to default and try to continue!\n");
> > > > > > +        info->organization.pagesize = 2048;
> > > > > > +        info->organization.oobsize = 64;
> > > > > > +        info->nand_size = SZ_1G;
> > > > > 
> > > > > Is this worth it? READ ID is the most basic command, when this doesn't
> > > > > work I don't think there's a point in continuing.
> > > > 
> > > > We had a case with Samsung K9K8G08U0E when sometimes (reasons
> > > > not known) the
> > > > 5th byte returned 0xff instead of correct values, all other
> > > > returned values
> > > > were correct. In this case the device booted successfully
> > > > because of this
> > > > hook. Maybe a better solution would be to check only the 5th
> > > > byte for 0xff
> > > > value (5th byte is not supported from all NAND vendors), if this
> > > > is the case
> > > > set the NAND size to 1GB.
> > > > Would that make more sense?
> > > 
> > > The best would be to track the issue down and to fix it ;)
> > > It's not very nice to assume it in case of read id failures it's exactly
> > > that NAND type you have troubles with.
> > > Does it help to reset the NAND chip before reading the ID?
> > > 
> > > Sascha
> > > 
> > 
> > I totally agree with fixing the issue and will look into it a bit more.
> > I will reset the NAND before reading the ID in the same way as it is
> > done before reading parameter page. I will leave the testing on over
> > night (reading the ID failed in about 1 out of 500 boots) and report the
> > results tomorrow.
> > 
> 
> I guess resetting fixed this problem, I got 50.000 boots without any
> problems. Will fix this in v2 :).
> 
> Should I now completely abandon setting the NANDs memory to default values
> and just return error on failure?

Yes please.

> Something like this:
> 
> > 	/*
> > 	 * If ONFI is not supported or if it fails try to get NAND's info from
> > 	 * "READ ID" command.
> > 	 */
> > 	ret = mxs_nand_reset(info, databuf);
> > 	if (ret)
> > 		return ret;
> > 	pr_debug("Trying \"READ ID\" command...\n");
> > 	ret = mxs_nand_get_readid(info, databuf);
> > 	if (ret) {
> > 		pr_err("xloader supports only ONFI and generic \"READ ID\" " \
> > 			"supported NANDs\n");
> > 		return -1;
> > 	}
> 
> I'm asking this because NANDs ID command is manufacturer specific, some
> manufacturers follow generic values and other unfortunately don't. In
> barebox NAND driver decoding is done based on manufacturer. I guess we could
> do the same here but I don't want to expand the PBL part too much just for
> reading FCB, where all NAND's memory parameters are decoded from.

We can do that once we have to later. Decoding the ID into page/block
sizes should also be done generically and not part of the i.MX6 NAND
loader.  Maybe we could also utilize parts of the NAND layer instead of
adding new code.

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox