Re: [PATCH V3 00/11] Multiplatform BMIPS kernel

[Arnd Bergmann <arnd@arndb.de>]

On Sunday 23 November 2014 18:40:35 Kevin Cernekee wrote:

> V2->V3:
> 
>  - Omit the BMIPS updates that have already been accepted into Ralf's tree.
>    They are still needed, but not reposted.
> 
>  - Make USB endian swap options conditional on "if CPU_BIG_ENDIAN".
> 
>  - Remove board listing from Documentation/devicetree/bindings/mips/brcm/soc.txt
> 
>  - Remove legacy device autodetection and chip ID decoding.  Legacy
>    boards/bootloaders will be supported by selecting a single DTB file
>    to compile into the kernel.
> 
>  - Refactor quirks code to match against DT "compatible" strings, not chip IDs.
> 
>  - Fix CPU1 boot (missing DT node) on 6329.
> 
>  - Remove @0 / addressing properties on non-reg nodes.
> 
>  - Remove bogus "brcm,bmips" bus registration.
> 
>  - Move UBUS peripherals onto a "simple-bus" and set DMA ranges for this
>    bus on bcm3384_zephyr.
> 
>  - Fix base addresses on 6328/6368 for "periph_intc@10000020".
> 
>  - Change the MIPS_L1_CACHE_SHIFT calculation so as to minimize the impact
>    on other builds (like bcm63xx).

Looks nice to me overall, with the new way of handling the dtb passing,
this seems much more flexible, and I guess it can be turned into an
actual multiplatform build if there is ever a desire to do that.

> Re: dma-ranges
> 
> dma.c implements a minimal remapping scheme just for the current UBUS
> peripherals.  The remapping is global, and it isn't the same mapping
> needed for PCI(e).  A more comprehensive solution will be needed before
> PCI support can be added.
> 
> On chips OTHER than 3384, remapping is only required on PCI (not UBUS or
> "rdb").  Notably, BCM7445, an ARM platform currently supported upstream,
> doesn't require dma-ranges for non-PCI devices.
> 
> I am hoping we can piggyback on top of the ARM dma-ranges code once it
> is merged.  This will allow for eliminating my dma.c.

Sounds good.

	Arnd