Increase memory access page alignment checking to 8KB instead of 4KB

[athornton]

gdwnldsKSC/es40@dad191f

This just transplants an es40 commit that forces 8K page alignment.

This not only resolves the installation issues, but seems to fix the network-driver-dying problem. In (admittedly somewhat limited) testing, I'm still seeing occasional storms of things like

CPacketQueue(rx_queue):add() packet lost! Size = 60.. dst: ff-ff-ff-ff-ff-ff .. src: 54-af-97-01-ca-f4
CPacketQueue(rx_queue):add() packet lost! Size = 60.. dst: ff-ff-ff-ff-ff-ff .. src: 54-af-97-01-ca-f4
CPacketQueue(rx_queue):add() packet lost! Size = 60.. dst: ff-ff-ff-ff-ff-ff .. src: 54-af-97-01-ca-f4
CPacketQueue(rx_queue):add() packet lost! Size = 60.. dst: ff-ff-ff-ff-ff-ff .. src: 54-af-97-01-ca-f4
CPacketQueue(rx_queue):add() packet lost! Size = 60.. dst: 01-00-5e-00-00-01 .. src: fc-34-97-46-58-30

But where that used to be the point at which the network adapter died and you'd have to restart axpbox to get it working again, now it appears to be recovering. That can happen but if I telnet or FTP to the emulated system it's still responsive, which was not the case before the alignment patch.

At any rate...I think this means that, with this patch, you can use axpbox with a TCP/IP network pretty much indefinitely.

[RaymiiOrg]

Thanks! I've updated your comment and title, to link to the original commit (gdwnldsKSC/es40@dad191f) and to highlight the actual change. Also approved a github actions workflow run, but the rest of the merging is up to @lenticularis39

[lenticularis39]

We probably shouldn't always printf that, since it might slow down emulation if there is too much unaligned access, but that can be solved later.