[PATCH 1/5] Add the explanation and sample of RapidIO DTS sector to the document of booting-without-of.txt file.

Zhang Wei-r63237 Wei.Zhang at freescale.com
Wed Jun 13 19:37:37 EST 2007


Hi, Segher, 

> 
> >>> +    - device_type : Should be "rapidio"
> >>
> >> There is no OF binding, so no.
> >
> > So, we need to define it.
> 
> If you want to.  Until that has been done, don't use
> a "device_type".  Linux won't use it, anyway.

Do you have another ideas about that? Only remove it?

> 
> >>> +    - compatible : Should be "fsl,rapidio-v0.0" or
> >> "fsl,rapidio-v1.0"
> >>> +      and so on. The version number is got from IP Block Revision
> >>> +      Register of RapidIO controller.
> >>
> >> It's better to use real device names, just like everyone
> >> else.
> >
> > Some silicons of Freescale processor are the same RapidIO 
> controller,
> > such as mpc8540/mpc8560 are the same (v0.0), mpc8548/mpc8641 are the
> > same (v1.0). For v1.0 RapidIO controller, should we use mpc8548 or
> > mpc8641? Those will make people confused.
> 
> Not at all.  On an 8641 it could be
> 
> 	compatible = "fsl,mpc8641-rapidio" "fsl,mpc8548-rapidio";
> 
> which states "this is the 8641 thing and it is compatible
> to the 8548 thing".  Perfectly clear.
> 
> > Using IP Block Revision is a
> > clear choice.
> 
> I don't think so.  For one thing, it describes a version of
> a cell design, not a version of an actual device.  For another
> thing, if I hear "8641" I know what you're talking about (sort
> of, anyway), but I draw a blank stare if you say "v1.0".  I'm
> sure I'm not the only one.  Concrete names are good.
> 

>From the different view ways, there are different results. Getting the
version from RapidIO IP revision register is clear to me. :)

> >>> +    - #address-cells : Address representation for
> >> "rapidio" devices.
> >>> +      This field represents the number of cells needed 
> to represent
> >>> +      the RapidIO address of the registers.  For
> >> supporting more than
> >>> +      36-bits RapidIO address, this field should be <2>.
> >>
> >> More than 32 bit?
> >
> > Yes, RapidIO bus address width is 34 bits.
> 
> You said "more than 36 bit", I tried to ask if that is a typo
> perhaps.

Ya, caught by you! I'll fix it in next version. :)

> 
> >> No.  The format of an "interrupts" entry is defined by
> >> the interrupt domain this device sits in, not by the
> >> device itself.
> >>
> > Do you misunderstand the meaning of 'interrupts'?
> 
> Hahaha.  No, I don't misunderstand what the "interrupts" property
> means.  Perhaps you do?
> 
> > These interrupts is
> > issued from the RapidIO controller to the pic controller for tx, rx,
> > err, doorbell and message.
> 
> But the rapidio node doesn't know or care what the interrupts
> are connected to, and neither should it.  That's what the
> interrupt mapping recommended practice is for.
> 

There are no rapidio device in it. Doorbell, msg are all parts of
rapidio controller.
For example, 8641 rapidio controller have 2 msg unit: msg0 and msg1.
They are not rapidio devices. Each msg unit has the tx_irq and rx_irq.

> >>> For this sector, interrupts order should be
> >>> +      <err_irq bell_outb_irq bell_inb_irq msg1_tx_irq msg1_rx_irq
> >>> +      msg2_tx_irq msg2_rx_irq ... msgN_tx_irq msgN_rx_irq>.
> >>
> >> That's to be defined in the binding for your specific device,
> >> not in a more generic rapidio binding.
> >
> > These description is just for compatible="fsl,rapidio-v*.*" rapidio
> > controller.
> 
> Okay, good.  Please make that way more obvious then :-)
> 
> >>> +		#address-cells = <2>;
> >>
> >> You want a #size-cells as well.
> >
> > The size is not used in this sector, so no defined.
> 
> The size _is_ used; in the "ranges" property in this node,
> for example.  It is also needed to describe the "reg" for
> any child node of this node.
> 
> A non-existant "#size-cells" means 1, and "#address-cells"
> means 2, so in principle you could do without these
> properties; but Linux doesn't parse the tree correctly in
> that case (which reminds me, I have some more patches to
> send).
> 

Ok, I'll add it in the next version for more religious.

Thanks!
Wei.



More information about the Linuxppc-dev mailing list