by Albert ARIBAUD

[permalink] [raw]

Subject: Re: [PATCH v8 1/5] mtd: nand: vf610_nfc: Freescale NFC for VF610, MPC5125 and others

Hi Stefan,

Le Thu, 30 Jul 2015 19:00:38 +0200, Stefan Agner <[email protected]> a
écrit :

> Hi Albert,
>
> On 2015-07-30 18:13, Albert ARIBAUD wrote:
> > Hi Stefan,
> >
> > Le Mon, 27 Jul 2015 18:42:41 +0200, Stefan Agner <[email protected]> a
> > écrit :
> >
> >> This driver supports Freescale NFC (NAND flash controller) found on
> >> Vybrid (VF610), MPC5125, MCF54418 and Kinetis K70. The driver has
> >> been tested on 8-bit and 16-bit NAND interface and supports ONFI
> >> parameter page reading.
> >>
> >> [...]
> >>
> >> diff --git a/drivers/mtd/nand/vf610_nfc.c b/drivers/mtd/nand/vf610_nfc.c
> >> new file mode 100644
> >> index 0000000..0da500e
> >> --- /dev/null
> >> +++ b/drivers/mtd/nand/vf610_nfc.c
> >> @@ -0,0 +1,640 @@
> >> [...]
> >
> > ... about line 708:
> >
> >> + err = devm_request_irq(nfc->dev, irq, vf610_nfc_irq, 0, DRV_NAME, mtd);
> >> + if (err) {
> >> + dev_err(nfc->dev, "Error requesting IRQ!\n");
> >> + goto error;
> >> + }
> >> +
> >> + vf610_nfc_init_controller(nfc);
> >
> > The call above is too early: vf610_nfc_init_controller() will test
> > for (nfc->chip.options & NAND_BUSWIDTH_16) but this option bit is only
> > set once nand_scan_ident() below has been run.
> >
> > This has the effect that even when the DT node specifies a 16-bit wide
> > bus, the controller is configured for 8-bit mode at this point, which of
> > course causes read failures. I've experienced this with a Vybrid SoC
> > and a Micron MT29F4G16ABADAH4 16-bit NAND.
> >
> >> + /* first scan to find the device and get the page size */
> >> + if (nand_scan_ident(mtd, 1, NULL)) {
> >> + err = -ENXIO;
> >> + goto error;
> >> + }
> >
> > Placing the call to vf610_nfc_init_controller() here, after the call
> > to nand_scan_ident() rather than before it, fixed the issue for me.
>
> Hm, since nand_scan_ident access the devices we actually want the
> controller initialized before we access it the first time. In most
> cases, the boot loader/boot ROM probably initialized the controller in a
> way that identifying the chip should work non the less. However, the
> safe way would be to initialize it before calling nand_scan_ident.

Correct -- there is a mutual dependency here between nand_scan_ident()
and vf610_nfc_init_controller(). Anyway, my call order switch was only
a hack to get things working.

> However, I see your point regarding bus width: With the change to
> nand_dt_init, we have that information after nand_scan_ident. There is
> actually more: Also the HW ECC settings are not yet parsed at that
> point, hence the ECC status and offset will not be initialized right.
>
> We could call the whole initialization twice. This would configure 8-Bit
> mode for the 16-Bit devices, but during initialization this is anyway
> the required default (ONFI). Or we split it up and call it something
> like vf610_nfc_preinit_controller and vf610_nfc_init_controller.
>
> What do you think?

I think creating a vf610_nfc_preinit_controller() function with no
dependency on the DT is the cleaner option.

> --
> Stefan

Cordialement,
Albert ARIBAUD
3ADEV