[Sound-open-firmware] SOF: Where is the purpose of host_msg ?

[Liam Girdwood]

On Wed, 2019-01-23 at 16:42 +0200, Daniel Baluta wrote:
> Please s/Where/What/ in the title. :))
> 
> On Wed, Jan 23, 2019 at 4:41 PM Daniel Baluta <daniel.baluta at gmail.com> wrote:
> > Hi Pierre/Liam,
> > 
> > In src/include/sof/ipc.h there is:
> > 
> > struct ipc {
> >      uint32_t host_msg;»     »       /* current message from host */
> > //
> > }
> > 
> > This keeps the message received from IA core. Anyhow, DSP core
> > doesn't seem to use it at all.
> > 

Yeah, this is legacy that could be removed.

> > Except maybe for some logging:
> > 
> > src/drivers/intel/baytrail/ipc.c:                       _ipc->host_msg =
> > msg;
> > src/drivers/intel/baytrail/ipc.c:       trace_ipc("ipc: msg rx ->
> > 0x%x", ipc->host_msg);
> > src/drivers/intel/cavs/ipc.c:                   _ipc->host_msg = msg;
> > src/drivers/intel/cavs/ipc.c:   trace_ipc("ipc: msg rx -> 0x%x", ipc-
> > >host_msg);
> > src/drivers/intel/cavs/sue-ipc.c:       trace_ipc("ipc: msg rx ->
> > 0x%x", ipc->host_msg);
> > src/drivers/intel/haswell/ipc.c:                        _ipc->host_msg =
> > msg;
> > src/drivers/intel/haswell/ipc.c:        trace_ipc("ipc: msg rx ->
> > 0x%x", ipc->host_msg);
> > src/include/sof/ipc.h:  uint32_t host_msg;              /* current
> > message from host */
> > 
> > I am asking this because initially I thought the communication between
> > IA core and DSP involves:
> > 
> > - sending cmd using shim layer
> > - sending actual data using mailbox.

Yep, on the Intel MMIO based architecture we can think of the SHIM as the
doorbell and shared memory as the mailbox. The SHIM doorbells can assert IRQs on
host and DSP (one doorbell for host -> dsp and the other for DSP -> host)

> > 
> > But it seems that all the information is exchanged via the mailbox memory.
> > So,
> > going back to my question will host_msg be used in the future?
> > 

Yes, this was changed as mailbox was a more architecture neutral method of IPC
i.e. it did not depend on passing metadata via special registers.

> > If not, I think it is best to remove it because it is confusing.

Feel free to send PR and remove :)

> > 
> > The same happens when DSP tries to send something to IA.
> > 
> > For example it writes msg->header. But this is never used on the other side.
> > 
> > ipc_platform_send_msg
> > 
> > »       /* now interrupt host to tell it we have message sent */
> > »       shim_write(SHIM_IPCDL, msg->header);
> > »       shim_write(SHIM_IPCDH, SHIM_IPCDH_BUSY);
> > 
> > Also, would be easier for me on the ARM side to only send info via mailbox
> > and
> > notification via what you call shim layer.

Also makes it easier for SPI and virtio too.

Thanks

Liam

> > 
> > thanks,
> > Daniel.