Re: TCP Segmentation Offloading (TSO)

[Paul Mackerras <paulus@samba.org>]

David S. Miller writes:

> It is real requirement, x86 works because unaligned
> access is handled transparently by cpu.
> 
> But on every non-x86 csum_partial I have examined, worse than
> 2-byte aligned data start is totally not handled.  It is not difficult
> to figure out why this is the case, everyone has copied by example. :-)

PPC and PPC64 are OK, I believe, since the CPU handles (almost) all
unaligned accesses in hardware.  (Some PowerPC implementations trap if
the access crosses a page boundary but the newer ones even handle that
case in hardware, and if we do get the trap we fix it up.)

I notice though that if the length is odd, we (PPC) put the last byte
in the left-hand (most significant) byte of a 16-bit halfword, with
zero in the other byte, and add it in, whereas i386 puts the last byte
in the least-significant position.  Hmmm... should be OK though since
I presume the result will be reduced and then converted to network
byte order before being put in the packet.  And since there is an
end-around carry we should end up with the same bytes that i386 does.

Paul.