On Wed, Oct 4, 2017 at 4:32 AM, Sudeep Holla <sudeep.holla@xxxxxxx> wrote: > > > On 04/10/17 12:24, Arnd Bergmann wrote: >> On Thu, Sep 28, 2017 at 3:11 PM, Sudeep Holla <sudeep.holla@xxxxxxx> wrote: >>> Some of the mailbox controller expects controller specific data in order >>> to implement simple doorbell mechanism as expected by SCMI specification. >>> >>> This patch creates a shim layer to abstract the mailbox interface so >>> that it can support any mailbox controller. It also provides default >>> implementation which maps to standard mailbox client APIs, so that >>> controllers implementing doorbell mechanism need not require any >>> additional layer. >>> >>> Cc: Arnd Bergmann <arnd@xxxxxxxx> >>> Signed-off-by: Sudeep Holla <sudeep.holla@xxxxxxx> >> >> Another level? Now we have three levels of stacked mailboxes, with >> the highest level being the combined mailbox/memory, then the shim, >> and below it the hardware mailbox. >> >> Can you try to come up with a way to do this with fewer abstractions? >> > > I completely agree with you. I was against this but Jassi recommended > this. I just wanted this SCMI to work with mailbox controllers that > support simple doorbell mechanism as specified in the specification but > Jassi disagrees with that. > >> Maybe you could assume that the mailbox itself can take variable-length >> data packets, and then use the shim here for those that require >> something else? >> > > As per SCMI specification, we pass all the data in shared memory and it > just expects to use a simple doorbell feature from hardware mailbox > controllers. It's done that way intentionally to avoid dependency on h/w > and we for sure will have variety of it and that defeats the purpose > of this standard specification. > > Also, I have added shim only for specific controllers that need them. > E.g. ARM MHU as Jassi disagreed to add doorbell mechanism to that. > mbox_if provides default implementation that just calls direct mailbox > APIs. > drivers/mailbox is a framework for interfacing/abstracting hardware mailboxes. If you're starting to layer mailboxes ontop of each-other chances are very high that you're confusing it with the computer science term "mailbox". Abstracting a doorbell-like piece of hardware behind the mbox framework makes a lot of sense, but the interface between your clients and the code that fills out shared memory and then invokes said doorbell is a higher level of "mailbox" and is probably better implemented using a direct function call. Regards, Bjorn -- 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
