[aspeed][Nexthop] Enable UDMA on the AST2700 BMC uart12 console

[jack-nexthop]

Enable dma-mode on uart12 in the NextHop B27 R0 device tree and add fixes to make the UDMA console usable:

-serial-8250_aspeed-fix-UDMA-ring-pointer-desync.patch:
Keep driver's xmit kfifo and RX ring indices aligned with the
UDMA engine's ring pointers (software readonly, preserved
across channel reset) across flush_buffer, set_termios
and open. Without this the console hangs in tcsetattr()
on the second and later logins and wedges RX on the
first termios change (bricking console until
serial-getty@ttyS12.service is restarted)

-soc-aspeed-udma-lower-idle-timeout.patch
Lower UDMA_TMOUT from 0x200 (~98.5ms on AST2700) to 0x40 (~12 ms)
so single keystrokes on a UDMA console are less visibly delayed

Verified on NextHop B27 R0 (AST2700) hardware

[mssonicbld]

/azp run

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[paulmenzel]

In the commit message, please don’t indent Verified on NextHop B27 R0 (AST2700) hardware.

Also, for people not knowing ASPEED code, on what repository are your patches based, and what is the plan to upstream them?

[mssonicbld]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).

[jack-nexthop]

In the commit message, please don’t indent Verified on NextHop B27 R0 (AST2700) hardware.

Also, for people not knowing ASPEED code, on what repository are your patches based, and what is the plan to upstream them?

Thanks, fixed the indentation.

These are original patches against AspeedTech-BMC/linux (branch aspeed-master-v6.12, apply cleanly to aspeed-master-v6.18 as well) and are not backports.

Aspeed has reproduced the issue on their side and is currently validating these patches for inclusion in their tree. Once they land there, SONiC picks them up on the next base bump and can drop the out-of-tree copies.

No other platform enables this channel, so this is a no-op outside NextHop B27 R0, where this fix has been verified.