On Tue, Jul 17, 2018 at 03:38:17PM +0200, Alexander Graf wrote: > On 07/17/2018 03:34 PM, Felix Brack wrote: > > > >On 17.07.2018 15:21, Tom Rini wrote: > >>On Tue, Jul 17, 2018 at 12:45:51PM +0000, Alexey Brodkin wrote: > >>>Hi Felix, > >>> > >>>>-----Original Message----- > >>>>From: Felix Brack [mailto:[email protected]] > >>>>Sent: Tuesday, July 17, 2018 3:13 PM > >>>>To: Alexander Graf <[email protected]>; Lokesh Vutla <[email protected]>; > >>>>[email protected] > >>>>Cc: Wolfgang Denk <[email protected]>; Tom Rini <[email protected]>; Marek > >>>>Vasut <[email protected]>; Patrice Chotard > >>>><[email protected]>; Michal Simek <[email protected]>; Simon > >>>>Glass <[email protected]>; Alexey Brodkin > >>>><[email protected]>; Bin Meng <[email protected]>; Ley Foon > >>>>Tan <[email protected]>; Patrick Delaunay > >>>><[email protected]>; Mario Six <[email protected]>; Stefan Roese > >>>><[email protected]>; Bernhard Messerklinger > >>>><[email protected]> > >>>>Subject: Re: [PATCH v2] serial: ns16550: Add register shift variable > >>>[snip] > >>>>Adding a separate PORT in ns16550_serial_ids for a particular > >>>>architecture, platform or SoC would be an option. However the patch I > >>>>posted is much more generic as it offers to set the reg-shift property > >>>>for no matter what architecture, platform or SoC. It can also easily be > >>>>extended by adding more conditional defaults to the Kconfig file. > >>>I'd say we're dealing with just one corner-case here. > >>>If I understand a concept of Device Tree it is supposed to describe your > >>>hardware. Thus if reg shift exists in your HW it should be explicitly > >>>mentioned in > >>>your .dts. If for some [historical] reason you have to deal with > >>>"incorrect" .dts then > >>>I'd prefer to have mentioned quirk with a separate PORT in > >>>ns16550_serial_ids > >>>instead of adding yet another Kconfig option. > >>So, this is part of the problem I suppose. I don't know _why_ we can't > >>just add the correct and valid reg-shift property to the dtsi file in > >>Linux and be done with it. Then the U-Boot driver would work because we > >>parse that property. > >> > >The only reason I can see why the <reg-shift> property "can't be added" > >to the Linux .dtsi file is that there is nothing broken in Linux. Hence > >we would actually ask Linux to add a property required by U-Boot. > > In the DT you can describe hardware specifics by either a different > compatible string or by additional properties on a generic compatible. So > > compatible = "ti,omap3-uart"; > > is correct, as is > > compatible = "ns16550"; > reg-shift = 2; > > There might be more that gets implied by the omap3-uart compatible that I'm > not aware of, but in a nutshell it's all about whether you use a generic > compatible string or a device specific one. Linux went for the specific one, > so it didn't need reg-shift. U-Boot (incorrectly) treats the device specific > compatible string as generic which is why you see the failure.
So, to answer my own questions, drivers/tty/serial/8250/8250_omap.c takes the compatible and forces the reg-shift. Honestly I assume they do this because there's other things being handled in those SoC-specific files in there. I guess we need to follow suit, sigh. -- Tom
signature.asc
Description: PGP signature
_______________________________________________ U-Boot mailing list [email protected] https://lists.denx.de/listinfo/u-boot
