ntb: Add Xeon Gen3 support

[bret_ketchum_dell.com]

Looks good to me.

sys/dev/ntb/ntb_hw/ntb_hw_intel.c

1301

Might want to keep the comment and assert on default.

[markj]

sys/dev/ntb/ntb_hw/ntb_hw_intel.c

1301

Note that the NTB type is an enum, so I'm not sure that having a default case makes sense.

1312

Isilon's patch and Linux and DPDK perform 8-byte reads and writes for doorbells.

[markj]

• A few fixups, switch on the NTB hardware type to make code a bit clearer.
• Update the man page.

[cem]

No objection. I don’t really want to be an NTB maintainer, fwiw. I skimmed maybe half of this and it seemed reasonable.

[markj]

In D26683#596653, @cem wrote:

No objection. I don’t really want to be an NTB maintainer, fwiw. I skimmed maybe half of this and it seemed reasonable.

Thanks for looking. If I don't get any feedback in the next few days I'll just go ahead and commit.

[cem]

In D26683#596676, @markj wrote:

Thanks for looking. If I don't get any feedback in the next few days I'll just go ahead and commit.

👌

[vangyzen]

I'm not an NTB expert, but LGTM.

sys/dev/ntb/ntb_hw/ntb_hw_intel.c

1146

desired_vectors was just assigned to num_vectors, so this is always true.

1155

Indentation?

1301

It would catch a scribbler earlier.

1835

Indentation?

2142–2143

Indentation?

I assume this is a rare case, but adding the sizes to the error message could be helpful.

3253–3256

You could remove the condition and keep only limit = base + size;. Or maybe a different condition is needed?

[markj]

sys/dev/ntb/ntb_hw/ntb_hw_intel.c

1155

I think this can just be removed.

3253–3256

I believe we can just remove the conditional.

[markj]

Address feedback.