GitLab

The newer linux drivers now use 16b instead of 32b descriptors. So far
this leads to an abort. This adds proper support for 16b descriptors.

Depending on the order some of the constructors get executed, instances of the
logger can end up with a null pointer for the runner. This led to a segfault
when calling TimePs() on that nullptr.

Here we fix this in two steps: first the logger takes a device pointer, which
in turn contains a pointer to the runner that will eventually be initialized.
Also fix up the logger to simply use 0 as a timestamp for log events before the
runner is there.

make lint should pass again now

As the python scripts pass the same arguments to both. We had dropped the sync
mode here but not in nicbm. For now just make it consistent

Currently sequence number is printed signed, while gem5 prints it unsigned as
it should be.

Previously we were calling completions etc. directly, but then the
prioritization does not match what gem5 does. This is slightly slower, but we
want to match gem5 exactly here.

Also splits out arg parsing from running so code can use the mac address in
initializationg before calling run.

Disabled by default as verilator is a bit tricky with versions etc.

Fix up behavioral nic simulators to compile and link again.

Previously the build was broken with debugging enabled.

No reason to print a warning on every drop. Slows down kvm simulation
quite a bit

We had this set to 64bit which is too small for 100G. Bumped it up to
512.

This drastically improves throughput for single-core hosts, as the
driver always polls all event queues on an interrupt, which involves at
least one MMIO read and multiple writes.

To support use-cases with multiple NIC interfaces.

With this nopaxos looks better.

Add explicit for single param constructors.