[multitop] Port DIF to multitop, port two tests to multitop

[pamaury]

Follow up PR on #25807, this handles the DIF and shows how to port two tests.

This PR also does one notable change to the DT which was discussed on #25807 and pushed to this PR: rename "signal" to "peripheral I/O" to match the CIP specification.

[jwnrt]

The code LGTM.

It looks like the compiler did pick up on the enum = int + 1 problem from the previous PR. I think the other CI errors are casting problems. LGTM once these are fixed and it passes CI.

[a-will]

We'll see if a single set of clkmgr DIFs can really hold up for multi-top. ipgen templated IPs may not all work with this sort of abstraction. For that matter, the OTP DIFs use named partitions in places, so that probably won't due...

Stuff to ponder for the future.

[pamaury]

Yes future will tell if a single DIF can handle that but it's definitely not the job of this PR to change the current approach ;)

[pamaury]

I have done a substantial overall of the PR:

• Now the pads and dt structures are access by indices, instead of pointers. Concretely, instead dt_pad_t and dt_<ip>_t become enums, like

/**
 * List of instances.
 */
typedef enum dt_uart {
  kDtUart0 = 0, /**<  */
  kDtUart1 = 1, /**<  */
  kDtUart2 = 2, /**<  */
  kDtUart3 = 3, /**<  */
  kDtUartCount = 4, /**< Number of instances */
} dt_uart_t;

and all APIs functions take a dt_uart_t as input. The actual struct and table is move to a dt_<ip>.c file and the types are made private. For the perspective of API users, the only change is that it's not a pointer type anymore.

• I added two functions to the pinmux DIF to make it easy to connect a MIO using the new dt_periph_io_t and dt_pad_t types.
• the dt/dt.h header is gone, everything is moved to either dt/dt_api.h or dt/dt_<ip>.h. This also means one needs to include one header per IP, like for the DIFs.
• the very large dt/dt.c is gone, it is now either moved to a new dt/dt_api.c or split into dt/dt_<ip>.c which makes more sense.

[pamaury]

So this PR is going to break Earlgrey whatever I do because the patching replaces top_earlgrey.h by the EB header but the DT headers are still generated for earlgrey and those use constants from top_earlgrey.h. We know that it is inevitable that EB breaks at some point and will need to be repaired when we can compile more code for EB.

I suggest we disable temporarily the EB CI tests for the time being? @vogelpi

[vogelpi]

So this PR is going to break Earlgrey whatever I do because the patching replaces top_earlgrey.h by the EB header but the DT headers are still generated for earlgrey and those use constants from top_earlgrey.h. We know that it is inevitable that EB breaks at some point and will need to be repaired when we can compile more code for EB.

I suggest we disable temporarily the EB CI tests for the time being? @vogelpi

SGTM, but let's make sure @a-will is in the loop as well as he used EB as a prototype, too.

[a-will]

Technically, these enums are still top-specific, since these are constructed from the instance names in the module section of the top hjson, right?

But we can adjust this whenever.

[pamaury]

Yes. The most logical thing to do is usually to go over all instances of this IP and test them one by one. In this case, test each aon_timer (even though it's almost sure that there is only one). Alternatively, we could just test the first one (set the index to 0).

[a-will]

At first glance, it doesn't look like this enum has a unique namespace. I'd wager that name collisions are unlikely, but we might want to adjust this at some point. The suffix is controlled by the user's hjson file, so it's less controlled. (We could choose to provide naming guidelines instead, though.)

[pamaury]

True, it's named dt_<something>_desc_t so technically could get a collision. We could alternatively name it dt_desc_<something>, this should be enough to avoid collisions. Side node: this type is local to the file so we could technically give it a fixed name, ie dt_this_ip_desc_t and it would also solve the problem.

[a-will]

So this PR is going to break Earlgrey whatever I do because the patching replaces top_earlgrey.h by the EB header but the DT headers are still generated for earlgrey and those use constants from top_earlgrey.h. We know that it is inevitable that EB breaks at some point and will need to be repaired when we can compile more code for EB.
I suggest we disable temporarily the EB CI tests for the time being? @vogelpi

SGTM, but let's make sure @a-will is in the loop as well as he used EB as a prototype, too.

Yes, this is fine. I hear @pamaury has already solved this in the multitop_dev branch, so it'll only be a temporary hiatus. 😄

stale