coresight: discovery: Avoid double offset in root component creation #1717
Conversation
Because the APAccessMemoryInterface includes and offset and the component factory API includes an address base, the base address offset at cpmid.address was being applied twice.
| @@ -262,7 +262,7 @@ def _create_root_component(self, cmpid): | |||
| memif = APAccessMemoryInterface(self.dp, ap_address) | |||
|
|
|||
| # Instantiate the component and attach to the target. | |||
| component = cmpid.factory(memif, cmpid, cmpid.address) | |||
| component = cmpid.factory(memif, cmpid, 0) | |||
There was a problem hiding this comment.
I tracked down this parametter to be set to CoreSightComponent._address.
So if I understand correctly, this should be set to None instead.
I can see that the AP Address is used as an offset in APAccessMemoryInterface. Could that be the issue ?
Are all read/write operations to this root AP meant to be relative to the base address?
EG, hypothetically, if there’s an SCS right on the root AP, then there could be some read/write in the 0xExxx_xxxx space even if the root’s base address is in 0xFxxx_xxxx, couldn’t it?
Could be please point to where it is applied a second time?
There was a problem hiding this comment.
I tracked down this parametter to be set to
CoreSightComponent._address.So if I understand correctly, this should be set to None instead.
None makes sense instead of 0 here.
I can see that the AP Address is used as an offset in
APAccessMemoryInterface. Could that be the issue ? Are all read/write operations to this root AP meant to be relative to the base address?EG, hypothetically, if there’s an SCS right on the root AP, then there could be some read/write in the 0xExxx_xxxx space even if the root’s base address is in 0xFxxx_xxxx, couldn’t it?
Could be please point to where it is applied a second time?
I was debugging an issue where a funnel was faulting on access: https://github.com/pyocd/pyOCD/blob/main/pyocd/coresight/funnel.py#L47
Since it's using the instantiated APAccessMemoryInterface which is applying cmpid.address as an offset to all accesses the self.address + TraceFunnel.DEVID results in a double offset.
I don't have enough of a high level view to understand which application of this makes sense. Perhaps instantiating APAccessMemoryIterface w/o an offset (or an offset derived from somewhere else ) makes sense?
There was a problem hiding this comment.
Since it's today instead of yesterday I can say that the chip I was debugging was the rp2350 (see #1718). The error ends up being ignored and you can flash, erase, reset the rp2350 fine.
Because the APAccessMemoryInterface includes and offset and the component factory API includes an address base, the base address offset at cpmid.address was being applied twice.