Re: [PATCH 2/6] Documentation: devicetree: add bindings to support ARM MHU subchannels

From: Sudeep Holla
Date: Tue May 09 2017 - 05:37:03 EST




On 08/05/17 18:52, Bjorn Andersson wrote:
> On Mon 08 May 10:07 PDT 2017, Sudeep Holla wrote:
>
>>
>>
>> On 08/05/17 17:46, Jassi Brar wrote:
>>> On Mon, May 8, 2017 at 9:40 PM, Rob Herring <robh@xxxxxxxxxx> wrote:
>>>> +Bjorn
>>>>
>>>> On Tue, May 02, 2017 at 02:55:49PM +0100, Sudeep Holla wrote:
>>>>> The ARM MHU has mechanism to assert interrupt signals to facilitate
>>>>> inter-processor message based communication. It drives the signal using
>>>>> a 32-bit register, with all 32-bits logically ORed together. It also
>>>>> enables software to set, clear and check the status of each of the bits
>>>>> of this register independently. Each bit of the register can be
>>>>> associated with a type of event that can contribute to raising the
>>>>> interrupt thereby allowing it to be used as independent subchannels.
>>>>>
>>>>> Since the first version of this binding can't support sub-channels,
>>>>> this patch extends the existing binding to support them.
>>>>>
>>>>> Cc: Alexey Klimov <alexey.klimov@xxxxxxx>
>>>>> Cc: Jassi Brar <jaswinder.singh@xxxxxxxxxx>
>>>>> Cc: Rob Herring <robh+dt@xxxxxxxxxx>
>>>>> Cc: devicetree@xxxxxxxxxxxxxxx
>>>>> Signed-off-by: Sudeep Holla <sudeep.holla@xxxxxxx>
>>>>> ---
>>>>> .../devicetree/bindings/mailbox/arm-mhu.txt | 44 ++++++++++++++++++++--
>>>>> 1 file changed, 41 insertions(+), 3 deletions(-)
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/mailbox/arm-mhu.txt b/Documentation/devicetree/bindings/mailbox/arm-mhu.txt
>>>>> index 4971f03f0b33..86a66f7918e2 100644
>>>>> --- a/Documentation/devicetree/bindings/mailbox/arm-mhu.txt
>>>>> +++ b/Documentation/devicetree/bindings/mailbox/arm-mhu.txt
>>>>> @@ -10,21 +10,40 @@ STAT register and the remote clears it after having read the data.
>>>>> The last channel is specified to be a 'Secure' resource, hence can't be
>>>>> used by Linux running NS.
>>>>>
>>>>> +The MHU drives the interrupt signal using a 32-bit register, with all
>>>>> +32-bits logically ORed together. It provides a set of registers to
>>>>> +enable software to set, clear and check the status of each of the bits
>>>>> +of this register independently. The use of 32 bits per interrupt line
>>>>> +enables software to provide more information about the source of the
>>>>> +interrupt. For example, each bit of the register can be associated with
>>>>> +a type of event that can contribute to raising the interrupt.
>>>>
>>>> Sounds like a doorbell? (i.e. a single bit mailbox). Bjorn is doing
>>>> something similar for QCom h/w. I guess the difference here is you have
>>>> 32 sources and 1 output. It seems to me these should be described
>>>> similarly.
>>>>
>>> Yes, QCom controller triggers different interrupt for each bit of a
>>> 32bits register i.e, each signal is associated with 1bit information.
>>> Whereas MHU signals 32bits at a time to the target cpu.
>>
>> Agreed. I had a look at Qcom driver, not entirely clear if each bit as
>> interrupt as I don't see any interrupt support there.
>
> Each of the (used) bits in the Qualcomm HW are routed to a interrupt
> controller in the remote processors.
>
> As the APCS IPC is one way and each incoming "channel" is exposed as a
> separate physical interrupt they are directly consumed by the higher
> levels as needed - hence there's no traces of interrupts in the APCS
> IPC.
>

Indeed, I followed the full thread later and came to same understanding.

>> Also, it just adds
>> all the 32 channels which I am trying to avoid as at-most 4-5 will be
>> used while we end up creating 64 channels.
>>
>
> In the Qualcomm platform I'm looking at right now 15 of the 32 bits are
> used by the local CPU, so the utilization isn't awesome but I didn't
> feel the waste was worth addressing in my case.
>

Make senses, but this controller in question was used on other platforms
too and hence I wanted to retain existing behavior intact without much
bloat of memory for channel allocation.

> You should be able to provide a custom of_xlate that hides the fact that
> your mailbox-space is sparse - i.e. only register the mailboxes you have
> but expose them with ids as expected by the clients.
>

Spot on, that's exactly what I have done :)

--
Regards,
Sudeep