[DRAFT] Adapt kmac_app_agent to dynamic app interface

[rswarbrick]

This DV change is designed to adapt the kmac_app_agent to a more sensible interface after the changes in #30143 were merged. I have split out several parts into separate PRs:

• [kmac,dv] Make kmac_app_item::byte_data_q unsigned #30391
• [keymgr,dv] Rename "keymgr.*_kmac_agent" to just "kmac_agent" #30392
• [kmac,dv] Rename the "user digest share" items #30393
• [kmac,dv] Rename kmac_app_intf to kmac_app_if #30394
• [kmac,dv] Pass req/rsp to kmac_app_if through inout ports #30395

Once those are merged the remaining commit is as follows:

[kmac,dv] Rewrite kmac_app_agent to be more general

This work adapts the agent to match the newly defined interface with
minimal changes. This should allow OTBN to interact with KMAC through
the interface in a more complicated way than existing users.

The agent that was in place isn't really able to handle the tweaked
protocol, so this commit rewrites it to allow extending things
smoothly.

Upsettingly, the commit is enormous! I couldn't figure out how to
split it up into smaller pieces. :-(

Important changes:

• The agent gets split in two ("kmac_app_device_agent" and
  "kmac_app_host_agent"). These have different sequencers,
  corresponding to the fact that they fundamentally drive different
  sorts of items.

• We no longer use push_pull_agent. It was already a slightly awkward
  fit (I think) and can't really support the new protocol.

• kmac_app_device_driver now watches app requests and creates
  sequence items to represent these requests. They get
  broadcast through m_req_port. That port is connected to the
  sequencer's m_req_fifo, which can be consumed by the sequence that
  sends responses.

• Sequence items are a bit more structured:

  • A request is sent in multiple beats, which each get a
    kmac_app_req_item where the last has m_last set.

  • The items representing the beats of a request are grouped into
    a "packet", represented by a kmac_app_req_packet_item. This
    contains a queue with the kmac_app_req_item objects for the
    beats.

  • A response is represented by a kmac_app_rsp_item.

  • The monitor sees complete transactions (a request followed by a
    response) and represents this as a kmac_app_mon_item, which
    contains a kmac_app_req_packet_item and a kmac_app_rsp_item.

• The signals in kmac_app_if are now driven through clocking blocks
  in both directions, depending on an if_mode variable.

• We don't use phase sequences any more. I think this is not really
  recommended and we can just start the "device responder" sequences
  from the test: probably a bit clearer. For example, see
  keymgr_base_test.sv (three lines of code).

Minor ("while we're at it") changes:

• It also exposes the request and response as inout ports, which
  makes it a bit easier to use. See hw/ip/keymgr/dv/tb.sv for an
  example of the change.

• Any new or dramatically-altered classes use out-of-block
  definitions and I've added meaningful documentation comments to
  functions, tasks and class variables.

• The imports in kmac_app_agent_pkg are a bit more precise (mainly
  because I was concentrating to make sure I got the right thing
  from the right place).

• The StrbAlignLSB_A assertions in kmac_app_intf were more
  complicated than needed. Group them into a single assertion. Also,
  express the assertions with default clocking/disable to make them
  easier to understand.

• Rather than using field macros for the new classes, I've defined
  do_print/do_copy manually. This is probably a bit more efficient at
  runtime and (more importantly) lets us be explicit about things
  like radixes. For example, kmac_app_req_item always uses
  hexadecimal for its m_data field.

• Since the strobe line is always required to be contiguous and
  aligned to the lsb, kmac_app_req_item represents it as just the
  number of valid bytes (m_num_bytes). This makes some of the code
  deailing with the item a bit simpler.

  For example, see kmac_app_host_seq: handling partial words when
  cfg.inject_zero_in_host_strb is now rather less mysterious.

• When porting the function from kmac_app_item, I renamed
  get_is_kmac_rsp_data_invalid to just is_kmac_rsp_data_invalid.
  There's no need for the "double accessor name".

• Since I was having to touch the code anyway, I've made the way
  scoreboards consume items from the agent a bit simpler. For
  example, keymgr_scoreboard now has two imp imports (for requests
  and transactions) and doesn't have to manually maintain fifos in
  the same way.

[etterli]

This first draft looks good! Most of the required changes to the other IPs can be found in the extra commits of #30143. But I assume you already know this :)

[rswarbrick]

Force-push rebases onto latest version of #30195

[rswarbrick]

I've just rebased this PR past the enormous changes that have landed (and tweaks to the DV code from the design team). I'm going to allow this to run through CI and then will review/hack on the code at this end a little bit until I'm convinced it's ready for review.

[etterli]

In case you are looking for the initial smoke test of the dynamic app interface, you can find it on this branch: https://github.com/etterli/opentitan/tree/otbn-kmac-if-combo

[rswarbrick]

This first force-push is prompted by trying to improve kmac_app_req_item, with the following changes:

• Implement do_compare for kmac_app_req_item
• Use print_field instead of print_field_int in kmac_app_req_item for
  fields that might be wider than 64 bits.
• Improve name for kmac_app_req_packet_item::get_share_byte_queue
• Make kmac_app_req_item::add_share_to_byte_queue use unsigned bytes
• Improve documentation around byte_length_c to explain when partial requests are ok.
• Drop a stray "unsigned" from m_last.
• Constrain m_delay not to be ridiculously large.
• Add m_using_masked_interface to kmac_app_host_seq to configure
  whether m_data_s1 can be nonzero, and whether partial beats are
  allowed in the middle of a burst.
• Allow termination to with an explicit empty message.
• Configure m_using_masked_interface when starting a kmac_app_host_seq
  from kmac_base_vseq.

[rswarbrick]

Changes in the latest force-push:

• Fix the silly signedness problems that caused CI failures last time
• Split out some of the changes in the big commit into precursors in the PR (to be reviewed and merged separately, I hope!)
• Fix fields missing from uvm_object_utils calls in
  kmac_app_agent_cfg. Rather than adding the fields to the macro
  calls, implement do_print, do_copy and do_compare.
• Fix grammar errors in description of zero_delays and rsp_digest_hs.
• Association fix in kmac_app_host_seq: I had written A && B || C when
  I meant A && (B || C).
• Fix a width mismatch in kmac_app_device_driver, which was using
  AppDigestW for a width of a signal that is actually MsgStrbW bits
  wide.

[rswarbrick]

Latest force-push pulls the change to the connections to kmac_app_if into a separate commit.

[rswarbrick]

Latest force-push addresses a few minor things, and also fixes a silly error in keymgr_env:

• Fix broken name in keymgr_env (I messed up a rebase and undid a
  rename from an earlier commit)

• Fix description of the DoneAssertAfterLast_A assertion in
  kmac_app_if. This is talking about a "done" signal, which has now
  been renamed to "rsp_valid".

• Improve radix for printing various fields in kmac_app_req_item and
  kmac_app_rsp_item.

• Add do_compare for kmac_app_rsp_item and add a constraint on the
  delay between request and response.

• Add do_compare to kmac_app_req_packet_item