Re: [PATCH v2] mtd: rawnand: xway: No hardcoded ECC engine, use device tree setting

[Miquel Raynal <miquel.raynal@bootlin.com>]

Kestrel seventyfour <kestrelseventyfour@gmail.com> wrote on Wed, 25 Aug
2021 10:47:40 +0200:

> Hi Miquèl,
> 
> Am Di., 24. Aug. 2021 um 19:22 Uhr schrieb Miquel Raynal
> <miquel.raynal@bootlin.com>:
> >
> > Hello,
> >
> > Kestrel seventyfour <kestrelseventyfour@gmail.com> wrote on Tue, 24 Aug
> > 2021 09:15:49 +0200:
> >  
> > > Hi Miquèl,
> > >
> > > Am Mo., 23. Aug. 2021 um 17:24 Uhr schrieb Miquel Raynal
> > > <miquel.raynal@bootlin.com>:  
> > > >
> > > > Hi Kestrel,
> > > >
> > > > Kestrel seventyfour <kestrelseventyfour@gmail.com> wrote on Mon, 23 Aug
> > > > 2021 13:19:43 +0200:
> > > >  
> > > > > Hi Miquèl,
> > > > >
> > > > > Am Do., 19. Aug. 2021 um 10:03 Uhr schrieb Miquel Raynal
> > > > > <miquel.raynal@bootlin.com>:  
> > > ...  
> > > > >
> > > > > thank you for your response.
> > > > > If I remove the nand-ecc-xxx properties in the device tree, the device with
> > > > > the Toshiba NAND chip is working. However, the device with the Micron
> > > > > NAND fails with NO ECC functions supplied; hardware ECC not possible,
> > > > > seems to be at line 5367 or equivalent.
> > > > > https://elixir.bootlin.com/linux/latest/source/drivers/mtd/nand/raw/nand_base.c#L5367
> > > > >
> > > > > It looks like the micron nand driver supports on die only if its
> > > > > specified int the
> > > > > Device tree:
> > > > > https://elixir.bootlin.com/linux/latest/source/drivers/mtd/nand/raw/nand_micron.c#L511
> > > > > The Micron NAND driver probably needs to set the ECC type to ON DIE if the
> > > > > variable ondie contains the supported attribute?!  
> > > >
> > > > You're right but I don't see any easy upstream-able solution here.
> > > > Changing the behavior in the Xway driver would certainly break users,
> > > > changing the behavior in the Micron driver would certainly break even
> > > > more users. The root cause being an absence of proper description (the
> > > > integration changed). Honestly I feel stuck, maybe you can try to
> > > > register your device, if it fails, change the integration in the driver
> > > > (to an ondie ecc engine) then retry?
> > > >
> > > > Thanks,
> > > > Miquèl  
> > >
> > > Do you think adding something like below at the following location
> > > https://elixir.bootlin.com/linux/latest/source/drivers/mtd/nand/raw/xway_nand.c#L223
> > > would be upstreamable (with or without device tree property?)?
> > >
> > >         err = nand_scan(&data->chip, 1);
> > >         if (err /* && of_property_read_bool(np, "lantiq,retry-on-die") */) {
> > >                 data->chip.ecc.engine_type = NAND_ECC_ENGINE_TYPE_ON_DIE;
> > >                 err = nand_scan(&data->chip, 1);
> > >                 if (err) return err;
> > >         }
> > >
> > > It still throws the kernel warning on first try, but the second try then works.  
> >
> > Can you please remind me what is xway/lantiq/your setup/how public it
> > is/who's using this driver?
> >
> > Thanks,
> > Miquèl  
> 
> Its for Openwrt, I would like to add support for 3 more devices
> AVM fritzbox 3490/5490 and 7490. They all have varying NAND chips.
> I have initially created a PR to have my initial patch tested:
> https://github.com/openwrt/openwrt/pull/4426
> There is already one device supported which has two DTBs one for
> Micron and one for non Micron (3370), but its not very straight forward.
> Without having this issue solved, flashing those devices would be
> possibly having issues depending on NAND chip or the awkward
> workaround of flashing one image and if it does not boot, boot the
> other one. Without self soldered serial console, it would not very
> easy to figure out the NAND manufacturer.
> The AVM stock firmware is old kernel and does not use device
> tree for NAND, they just query all possible manufacturers and set
> up NAND based on manufacturer query.

But in this case can't you check the 'root' compatible against certain
values and and some kind of quirk in the ->attach() hook to update the
ECC engine to the right one?

Thanks,
Miquèl