Move bank instrumentation to bank.{read/write}

[mandolaerik]

No description provided.

[mandolaerik]

This makes sure instrumentation works also for device-internal accesses, which allows a clean migration of sideband banks to the transaction interface. This also moves the feature further away from the memop, which means:

• the reported size is deduced from byte enables, rather than taken from the transaction
• the reported initiator is harder to reach; our workaround is to convey it in a session variable back-door, which in some cases will cause an incorrect initiator to be reported in instrumentation callbacks after read/write if SIM_transaction_wait interacts with reentrancy.

[syssimics]

Verification #12409052: fail

[syssimics]

Verification #12409305: pass

[mandolaerik]

Trigger Jenkins
Jenkins

[mandolaerik]

Jenkins looks ok

[lwaern-intel]

Somewhat surprised this shuffling of calls to the _callback family doesn't break anything.

You don't think VP passing is mostly due to their own local copy-pasting of these methods?

[mandolaerik]

VP passing is probably due to their tests not relying much on bank instrumentation (in part because bank instrumentation don't work for sideband accesses before this PR).

I think their copied overrides will cause callbacks to be called twice, but things that do rely on bank instrumentation are likely on the form "wait for next access", where a double callback probably isn't disruptive.

[mandolaerik]

Here is the issue for fixing this in VP.

[mandolaerik]

Before merging this, I should also add a release note.

[mandolaerik]

I can see that the VP duplication issue is scheduled for fixing before nov 17, so we can consider waiting for that.

[mandolaerik]

This perhaps deserves a discussion, there are probably valid use cases for read/write overrides where this change will be a regression. I don't know what to do about that.

[syssimics]

Verification #12413771: pass