Hi Cristian, > Subject: Re: [PATCH 0/2] firmware: arm_scmi: add notification completion > channel > > On Tue, May 07, 2024 at 11:34:59PM +0800, Peng Fan (OSS) wrote: > > Per spec: > > Hi Peng, > > thanks for doing this. > > > Completion interrupts This transport supports polling or interrupt > > driven modes of communication. In interrupt mode, when the callee > > completes processing a message, it raises an interrupt to the caller. > > Hardware support for completion interrupts is optional. > > > > i.MX95 SCMI firmware is fully interrupt driven, so Platform expects > > completion interrupt for Platform to Agent(P2A) notifictions. > > > > Add another optional mailbox channel for Agent to notify Platform that > > notification message has been accepted > > > > After notification channel status become freed, Agent will use the new > > mailbox channel to send completion interrupt to Platform. > > > > Add shmem_channel_intr_enabled to check channel flags. > > Glancing through the series I think the bindings and the code are quite good > and sensible to achieve the addition of the optional P2A completion interrupt. > > Your current solution, though, addresses only the case of a mailbox controller > providing unidirectional channels. (which I suppose is the one you are using > on your platform) Right. i.MX95 use unidirectional channels. > > We should take care to add such optional P2A completion interrupt support > also for the case of mailboxes sporting bidirectional channels. > > For bidirectional channels we really have already the needed bindings... > ...no changes there...you have just to also ring the doorbell on that same P2A > if completion IRQs are requested. > > IOW, when the DT description is made by 2 mboxes + 2 shmem, means we > have both A2P and P2A provided via bidirectional channels...we currently just > use the P2A (RX) channel to receive notifs/dresp via rx_callback (so using only > the platform_to_agent direction)...now we should also ring the P2A doorbell > on the existing P2A channel if the FLAG_INTR_ENABLED is set to signal the > completion interrupt on the other direction (agent_to_platform) > > Not sure if I have been clear or make it worst with my explanation :P Thanks for detailed explanation, it is clear enough. I will update to support bidirectional channel in V2. Thanks, Peng. > > Thanks, > Cristian
