On 06/29/2016 11:28 PM, Stephen Warren wrote:
On 06/28/2016 11:56 PM, Joseph Lo wrote:
On 06/29/2016 03:08 AM, Stephen Warren wrote:
On 06/28/2016 03:15 AM, Joseph Lo wrote:
On 06/27/2016 11:55 PM, Stephen Warren wrote:
On 06/27/2016 03:02 AM, Joseph Lo wrote:
snip.
Currently the usage of HSP HW in the downstream kernel is something
like
the model below.
remote_processor_A-\
remote_processor_B--->hsp@1000 (doorbell func) <-> host CPU
remote_processor_C-/
remote_processor_D -> hsp@2000 (shared mailbox) <-> CPU
remote_processor_E -> hsp@3000 (shared mailbox) <-> CPU
I am thinking if we can just add the appropriate compatible strings for
it to replace "nvidia,tegra186-hsp". e.g.
"nvidia,tegra186-hsp-doorbell"
and "nvidia,tegra186-hsp-sharedmailbox". So the driver can probe and
initialize correctly depend on the compatible property. How do you
think
about it? Is this the same as the (b) you mentioned above?
Yes, that would be (b) above.
However, please do note (a): I expect that splitting things up will turn
out to be a mistake, as it has for other HW modules in the past. I would
far rather see a single hsp node in DT, since there is a single HSP
block in HW. Sure that block has multiple sub-functions. However, there
is common logic that affects all of those sub-functions and binds
everything into a single HW module. If you represent the HW module using
multiple different DT nodes, it will be hard to correctly represent that
common logic. Conversely, I see no real advantage to splitting up the DT
node. I strongly believe we should have a single "hsp" node in DT.
We have 6 HSP block in HW. FYI.
Yes, we have 6 /instances/ of the overall HSP block. Those should each
have their own node, since they're entirely separate modules, all
instances of the same configurable IP block.
Above, I was talking about the sub-blocks within each HSP instance,
which should all be represented into a single node per instance, for a
total of 6 DT nodes overall.
Yes.
So, one thing still concerns me is that the binding and driver still
can't work with multiple HSP sub-modules per HSP block. It only supports
one HSP module per HSP block right how. Although, I said it matches the
model that we are using in the downstream kernel. But I still concern if
we need to enable and work with multiple HSP modules per HSP block at
sometime in future, then the binding and driver need lots of change to
achieve that. And the binding is not back-ward compatible obviously.
So I want to revise it again.
#mbox-cells: should be 2.
The mobxes property in the client node should contain the phandle of the
HSP block, HSP sub-module ID and the specifier of the module.
Ex.
hsp_top0: hsp@1000 {
...
#mbox-cells = <2>;
};
clientA {
....
mboxes = <&hsp_top0 HSP_DOORBELL DB_MASTER_XXX>;
};
clientB {
...
mboxes = <&hsp_top0 HSP_SHARED_MAILBOX SM_MASTER_XXX>;
};
Stephen, How do you think of this change?
Thanks,
-Joseph
--
To unsubscribe from this list: send the line "unsubscribe linux-tegra" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
