svn commit: r304439 - head/sys/dev/usb/net

[Hans Petter Selasky]

On 08/19/16 11:22, YongHyeon PYUN wrote:
> On Fri, Aug 19, 2016 at 11:11:56AM +0200, Hans Petter Selasky wrote:
>> On 08/19/16 10:55, YongHyeon PYUN wrote:
>>> I think the order is right but it was not tested on big-endian
>>> systems.
>>
>> Hi,
>>
>> I'm pretty sure the ifdef is wrong, because you write the fields one at
>> a time, using htole32():
>>
>>                         txhdr.mss = 0;
>>                         txhdr.len = htole32(AXGE_TXBYTES(m->m_pkthdr.len));
>>
>> Big endian machines don't re-order variables like this.
>>
>> You should remove the #else part.
>
> Wouldn't USB stack pass txhdr structure without any
> modification? And controller want to see len (low 32bits address)
> first and then mss (high 32bits address). On big endian systems I
> guess this should be reversed in host memory layout.  This is so
> confusing so I could be wrong.

The USB stack passes TXHDR as-is and the host controller is byte 
oriented, not 64-bit word oriented. That's why the layout is the same as 
long as you assign per 32-bit field.

--HPS