[U-Boot] SPL Platdata howto?

Fri Jan 11 07:37:40 UTC 2019

On Thu, Jan 10, 2019 at 1:57 PM Simon Glass <sjg at chromium.org> wrote:
>
> Hi Simon,
>
> On Fri, 4 Jan 2019 at 00:15, Simon Goldschmidt
> <simon.k.r.goldschmidt at gmail.com> wrote:
> >
> > On Fri, Dec 21, 2018 at 10:20 PM Simon Goldschmidt
> > <simon.k.r.goldschmidt at gmail.com> wrote:
> > >
> > >
> > >
> > > Am Fr., 21. Dez. 2018, 22:16 hat Simon Glass <sjg at chromium.org> geschrieben:
> > >>
> > >> Hi Simon,
> > >>
> > >> On Thu, 20 Dec 2018 at 14:32, Simon Goldschmidt
> > >> <simon.k.r.goldschmidt at gmail.com> wrote:
> > >> >
> > >> > Am 20.12.2018 um 21:53 schrieb Simon Goldschmidt:
> > >> > > Am 20.12.2018 um 18:37 schrieb Simon Glass:
> > >> > >> Hi Simon,
> > >> > >>
> > >> > >> On Thu, 20 Dec 2018 at 08:03, Simon Goldschmidt
> > >> > >> <simon.k.r.goldschmidt at gmail.com> wrote:
> > >> > >>>
> > >> > >>> Am 20.12.2018 um 15:49 schrieb Simon Glass:
> > >> > >>>> Hi Simon,
> > >> > >>>>
> > >> > >>>> On Wed, 19 Dec 2018 at 14:06, Simon Goldschmidt
> > >> > >>>> <simon.k.r.goldschmidt at gmail.com> wrote:
> > >> > >>>>>
> > >> > >>>>> Hi,
> > >> > >>>>>
> > >> > >>>>> while searching for bytes to save in SPL in order to add FIT signature
> > >> > >>>>> handling, I am currently trying to get socfpga-gen5 to use OF_PLATDATA.
> > >> > >>>>>
> > >> > >>>>> To begin, I stripped down socfpga_socrates_defconfig to absolutely
> > >> > >>>>> nothing but serial drivers in SPL (with some modifications to the
> > >> > >>>>> Kconfig) and enabled DEBUG_UART to see what's going on.
> > >> > >>>>>
> > >> > >>>>> Now while this config runs OK with a dtb (it just won't boot as drivers
> > >> > >>>>> are missing -> "failed to boot from all boot devices"), it does not find
> > >> > >>>>> the serial driver after enabling OF_PLATDATA.
> > >> > >>>>>
> > >> > >>>>> So since serial_rockchip.c already uses OF_PLATDATA and is based on
> > >> > >>>>> ns16550 that my socfpga-gen5 platform is using: what do I have to do
> > >> > >>>>> besides enabling OF_PLATDATA to get this working?
> > >> > >>>>>
> > >> > >>>>> I just seems like uclass_first_device does not find any UCLASS_SERIAL
> > >> > >>>>> deivce when OF_PLATDATA is enabled.
> > >> > >>>>
> > >> > >>>> There is the of-plat.txt README.
> > >> > >>>
> > >> > >>> Yes, I should have mentioned I already read that and still had those
> > >> > >>> questions. Kconfig help says README.platdata though. We probably should
> > >> > >>> update that link.
> > >> > >>>
> > >> > >>>> Basically the dtoc tool creates U_BOOT_DEVICE() declarations and links
> > >> > >>>> them with SPL. These should show up in your image and therefore be
> > >> > >>>> bound. You can call dm_dump_all() in SPL to see what what devices are
> > >> > >>>> bound. I presume you are calling spl_init()?
> > >> > >>>>
> > >> > >>>> You can look at what dtoc produces. The example serial driver for
> > >> > >>>> Rockchip is serial_rockchip.c
> > >> > >>>
> > >> > >>> I saw that as an example (because I also have an ns16550 compatible on
> > >> > >>> my board) but couldn't figure out why it is not bound. By debugging
> > >> > >>> 'dm_scan_platdata', 'lists_bind_drivers' and 'device_bind_by_name', by
> > >> > >>> now I know the driver names don't match. That is something I did not get
> > >> > >>> just by reading of-plat.txt. I'll work on a patch to clarify that document.
> > >> > >>
> > >> > >> Yes I'd really appreciate some patches here. It is hard to know what
> > >> > >> people won't understand and this feature could really do with a more
> > >> > >> details docs or a walk-through.
> > >> > >>
> > >> > >>>
> > >> > >>> Right now, serial works. I had to add a new platform specific driver
> > >> > >>> just like serial_rockchip though. For DTS, we can pass multiple
> > >> > >>> 'compatible' strings, but for platdata, we have to create multiple
> > >> > >>> drivers. That's a bit strange when porting boards...
> > >> > >>
> > >> > >> Yes it is. I'm not sure how to solve that though. Probably dtoc can be
> > >> > >> made smarter. Ideally you only need one device of each uclass in SPL.
> > >> > >
> > >> > > Would it work to use the unchanged 'compatible' string for the '.name'
> > >> > > of U_BOOT_DEVICE generated by dtoc? Then the binding of such drivers
> > >> > > could change from comparing names to comparing to compatibles. That
> > >> > > would make it more DTS-like.
> > >> > >
> > >> > > Then, I think we could need some kind of fallback code for properties
> > >> > > that are optional in the DTS. Maybe we can create defines for each dtd
> > >> > > struct so that drivers can test the existence of dtd sturct fields using
> > >> > > #ifdef. [Given the special usage, I guess it's OK to assume that theses
> > >> > > structs are only used once per DTS so that we don't have to worry about
> > >> > > how to solve this for multiple occurrences with different optional
> > >> > > parameters?]
> > >> > >
> > >> > > Oh, and then I think dtb_platdata.py should create the dtd instances as
> > >> > > const. I'll prepare a patch for that.
> > >> > >
> > >> > >>
> > >> > >>>
> > >> > >>>>
> > >> > >>>>>
> > >> > >>>>> (And when answering this, keep in mind I need to get MMC and QSPI
> > >> > >>>>> drivers working with OF_PLATDATA - I already fixed compiler errors in
> > >> > >>>>> those, nothing more.)
> > >> > >>>>
> > >> > >>>> Yes MMC should be OK, but QSPI might be blazing a bit of a trail.
> > >> >
> > >> > Hmm, QSPI works as well when hacking the things that the driver wants to
> > >> > parse from subnode properties. However, I haven't found a way to access
> > >> > the platform data of the spi-flash from the driver.
> > >> >
> > >> > Maybe we need to somehow make subnodes available in the dt-platdata
> > >> > structs to make that work?
> > >>
> > >> There is support for phandles but not for parent relationships. I
> > >> suppose it would not be impossible to add that in dtoc with a 'parent'
> > >> pointer.
> > >
> > >
> > > SPI flash actually needs it the other way round. At least the cadence qspi driver I'm using checks for a subnode that describes the flash chip.
> > >
> > > I'll see if I can add that to dtoc.
> >
> > By now I have SPI successfully running with platdata by adding child
> > arrays to the platdata struct via dtoc.
> >
> > However, probing the flash chip is not found in 'spi_get_bus_and_cs'
> > and so the transfer falls back to 100 kHz, which is of course bad.
> > That code expects a udevice child under the spi udevice. Looks like
> > that needs more changes than just in dtoc?
> >
> > Did you have SPI running with platdata on any board, yet?
>
> No. The implementation does not deal well with parent/child
> relationships, as with buses, and in every case we need to make
> changes.

I had done those changes to dtoc, but as Marek seems so opposed to
getting this integrated for socfpga, I don't know if I'll continue down that
road...

> That said, spi_get_bus_and_cs() is only used if you don't have the SPI
> device in the DT.

Hmm, I thought the way that spl_spi.c uses the spi flash framework
would result in spi_get_bus_and_cs() being used. I'll need to check
that again.

Regards,
Simon