Re: [Xen-devel] [PATCH v4 1/1] Support Odroid-XU board (Exynos 5410)

[Suriyan Ramasami <suriyan.r@xxxxxxxxx>]

On Wed, Sep 3, 2014 at 3:03 AM, Ian Campbell <Ian.Campbell@xxxxxxxxxx> wrote:
> On Wed, 2014-08-27 at 12:27 -0700, Suriyan Ramasami wrote:
>> On Tue, Aug 26, 2014 at 2:01 PM, Ian Campbell <ian.campbell@xxxxxxxxxx> 
>> wrote:
>> > On Tue, 2014-08-19 at 16:19 -0700, Suriyan Ramasami wrote:
>> >> +static int __init exynos5250_smp_init(void)
>> >> +{
>> >> +    return exynos_smp_init(EXYNOS5250_PA_SYSRAM);
>> >> +}
>> >> +
>> >> +static int __init exynos5_smp_init(void)
>> >> +{
>> >> +    struct dt_device_node *node;
>> >> +    u64 sysram_ns_base_addr;
>> >> +    u64 size;
>> >> +    int rc;
>> >> +
>> >> +    node = dt_find_compatible_node(NULL, NULL, 
>> >> "samsung,exynos4210-sysram-ns");
>> >
>> >
>> > Looking at Documentation/devicetree/bindings/arm/exynos/smp-sysram.txt
>> > in the Linx tree it seems that this node is supposed to be present on
>> > all systems, including 5250 and it seems to be present in all the DTBs I
>> > can see.
>> >
>> > IOW I think exynos5_smp_init and exynos5250_smp_init can just be
>> > exynos5_smp_init (perhaps even folding in exynos_smp_init too).
>> >
>> It definitely can be folded in, but as you have mentioned later on in
>> this email that Exynos5250 will stop booting cause of the erroneous
>> value of 0x204f000 instead of 0x2020000. If this is folded in, then
>> the arndale DT should be possibly corrected for it to play well with
>> XEN.
>
> I think this is already done, looking at v3.16-rc6 (just because I had
> it checked out) arch/arm/boot/dts/exynos5250.dtsi has:
>         sysram@02020000 {
>                 compatible = "mmio-sram";
>                 reg = <0x02020000 0x30000>;
>                 #address-cells = <1>;
>                 #size-cells = <1>;
>                 ranges = <0 0x02020000 0x30000>;
>
>                 smp-sysram@0 {
>                         compatible = "samsung,exynos4210-sysram";
>                         reg = <0x0 0x1000>;
>                 };
>
>                 smp-sysram@2f000 {
>                         compatible = "samsung,exynos4210-sysram-ns";
>                         reg = <0x2f000 0x1000>;
>                 };
>         };
>

I am a little confused here. We are using - compatible =
"samsung,exynos4210-sysram-ns" to pull the value for
sysram_ns_base_addr in function exynos5_smp_init(). I believe with the
exynos5250 DT you have quoted above, wouldn't that add up to 2020000 +
2f000? (which is not what we want for the arndale?) I shall wait on
your thoughts on this.

On the same train of thought, it would seem that EXYNOS5_MCT_BASE
should also be extracted from the DT (in function exynos5_init_time).

Also, I realized that the PLATFORM_START for the exynos5410 is thus:
PLATFORM_START(exynos5, "SAMSUNG EXYNOS5")
I believe it should instead be:
PLATFORM_START(exynos5410, "SAMSUNG EXYNOS5410")

I shall send out a patch version 5 with these changes.

>> > I'm unsure whether we should fallback to the existing PA_SYSRAM value,
>> > but my initial feeling is that we can get away without.
>> >
>> > My only concern is that right now EXYNOS5250_PA_SYSRAM == 0x02020000 but
>> > the DTB contains:
>> >
>> >         sysram@02020000 {
>> > [...]
>> >                 smp-sysram@0 {
>> > [...]
>> >                 smp-sysram@2f000 {
>> >                         compatible = "samsung,exynos4210-sysram-ns";
>> >                         reg = <0x2f000 0x1000>;
>> >                 };
>> >         };
>> >
>> > IOW I think the final address for the NS sysram will be 0x204f000 and
>> > not 0x2020000. I'm confused how this work{s,ed}. since the DT seem to
>> > suggest that the sysram at 0x2020000 is secure mode, so we can't be
>> > using it.
>> >
>> I was going over linux kernels (not mainline but vendor modified)
>> 3.10/3.13/3.14 and I find that though they have a DT, they do not seem
>> to use it entirely. Especially the Exynos modified ones. They have
>> many hardcoded values which they pull out of thin air and not use the
>> DT ones. I am guessing that is the reason they are working. I wonder
>> if the linux mainline 3.16 or 3.17 which I believe has support for the
>> arndale exynos5250 boots up because of the erroneous (or so I think)
>> sysram-ns value?
>
> I've not tried something more modern on an arndale but I'm told they are
> now multiplatform capable, which implies most stuff is correctly driven
> by DT, proof will be in the pudding though!
>
> Ian.
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel