[RFC] mailbox: Add Broadcom BCM2835 mailbox driver

Stephen Warren swarren at wwwdotorg.org
Tue Aug 20 23:30:07 EDT 2013

On 08/19/2013 01:28 AM, Craig McGeachie wrote:
> On 08/17/2013 11:47 PM, Craig McGeachie wrote:
>> I think Lubomir's implementation, with the atomic write/read interface,
>> might be the safest, but that it gains safety at the cost of
>> performance. I say performance a bit tongue in cheek because I don't see
>> the need for that many mailbox calls.  The only place where I thought it
>> might be an issue, vchiq, makes the call once to set an alternative call
>> mechanism.  Unless somebody is checking the chip temperature every 5us,
>> I can't see it being an issue.
> I take it all back.  Lubomir's implementation looks close to ideal. With
> the following caveats.
... [discussion of caveats]
> But I still don't think it should be in drivers/mailbox or make use of
> include/linux/mailbox.h.

That's unfortunate. If there's a mailbox sub-system, it really ought to
be suitable for plugging in arbitrary mailbox implementations, or
there's not really much point pretending like it's a subsystem...

You probably want to start a thread with Rob Herring and/or Suman Anna
re: what abstraction drivers/mailbox should be exposing, since they're
the two committers to that directory.

> How could this patch be progressed?  Is there anything I could do?

Per Documentation/SubmittingPatches, assuming Lubomir's patches are
already appropriately licensed and signed-off, you're quite free to take
those patches, modify them as necessary, make a note of what you
changed, add your sign-off, and repost the result for inclusion upstream.

More information about the linux-rpi-kernel mailing list