On 03/11/2015 12:32 PM, Tony Lindgren wrote:
> * Suman Anna <s-anna@xxxxxx> [150311 10:18]:
>> On 03/11/2015 11:26 AM, Tony Lindgren wrote:
>>> * Dave Gerlach <d-gerlach@xxxxxx> [150310 12:55]:
>>>> Suman and I have been looking at this together, so I can comment here. An
>>>> implementation like this is what Suman is referring to:
>>>>
>>>> + l4_wkup: l4_wkup@44c00000 {
>>>> + compatible = "am335-l4-wkup", "simple-bus";
>>>> + #address-cells = <2>;
>>>> + #size-cells = <1>;
>>>> + ranges = <0 0 0x44c00000 0x100000>,
>>>> + <1 0x0 0x44d00000 0x4000>,
>>>> + <2 0x80000 0x44d80000 0x2000>;
>>
>> Actually, this would be slightly different, something like
>> + ranges = <0 0 0x44c00000 0x100000>,
>> + <1 0 0x44d00000 0x100000>,
>> + <2 0 0x44e00000 0x4000>,
>> + <3 0 0x44e10000 0x2000>;
>>
>> and the M3 DMEM entry below will be adjusted as <1 0x80000 0x2000>.
>>
>>>> +
>>>> + wkup_m3: wkup_m3@1,0 {
>>>> + compatible = "ti,am3353-wkup-m3";
>>>> + reg = <1 0x0 0x4000>, /* M3 UMEM */
>>>> + <2 0x80000 0x2000>; /* M3 DMEM */
>>>> +
>>>> + ti,hwmods = "wkup_m3";
>>>> + ti,pm-firmware = "am335x-pm-firmware.elf";
>>>> + };
>>>> + };
>>>> +
>>>>
>>>> The of_* layer automatically translates everything so the pdata-quirks can still
>>>> match based on wkup_m3@44d00000. The existing wkup_m3_rproc driver works almost
>>>> entirely as is with this, all cpu addresses are read and mapped correctly but
>>>> the driver no longer will read the actual device addresses correctly which we
>>>> need for understanding where to load the firmware sections.
>>>
>>> OK. I still don't quite understand how these additional ranges make sense
>>> for other drivers connected to the l4_wkup. For wkup_m3, it makes sense if
>>> it allows you to translate directly to the m3 address space, but is that
>>> really the case here? Maybe you should have the ranges in wkup_m3 instead
>>> if you want addresses for the m3?
>>
>> The idea is to introduce an additional address element (first cell in
>> ranges) so that the immediate child nodes bus address is referenced as 0
>> (second cell) for translation for their child nodes. This is the
>> approach used by the current scm node in Tero's series for OMAP4+. This
>> will work tomorrow if we move the prcm, scrm node under l4_wkup with
>> changes only in those nodes, and have their child nodes reg properties
>> unchanged. I guess you can see the difference between the following two
>> patches from Tero's PRCM series,
>> https://patchwork.kernel.org/patch/5882831/ &
>> https://patchwork.kernel.org/patch/5882841/
>
> Well I just commented on Tero on that regarding the dra7 patch. I think
> we need to have separate scm instances for scm_device, scm_core and
> scm_wkup instead of doing multiple ranges. This based on looking at for
> example 5432 TRM "Figure 18-1. Control Module Overview".
>
> But here I think it's a different issue. You want to use ranges for getting
> the m3 address space for the firmware? I'm not convinced we should
> complicate the ranges for all l4_wkup drivers because of that.
Yes, to some extent that's true, as we want to compute the m3 address
space using the regs property. In anycase, the wkupm3 driver has to be
updated for both approaches, the existing patch wouldn't work as is with
the above convention of using 2 address cells. Since the l4_wkup node
would be added for the first time, we are trying to see what would be
the good approach w.r.t subsequent changes in the DTS if and when we
have to move other nodes to be under l4_wkup.
regards
Suman
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
