RE: [PATCH net-next 5/5] net: txgbe: support getting module EEPROM by page

["Jiawen Wu" <jiawenwu@trustnetic.com>]

> >>> +
> >>> +	for (i = 0; i < dword_len; i++) {
> >>> +		value = rd32a(wx, WX_FW2SW_MBOX, i);
> >>> +		le32_to_cpus(&value);
> >>> +
> >>> +		memcpy(&local_data[i * 4], &value, 4);
> >>> +	}
> >>
> >> the logic here is not clear from the first read of the code. effectively
> >> in the reply you have the same txgbe_hic_i2c_read struct but without
> >> data field, which is obviously VLA, but then you simply skip the result
> >> of read of txgbe_hic_i2c_read and only provide the real data back to the
> >> caller. Maybe you can organize the code the way it can avoid double copying?
> >
> > Because the length of real data is variable, now it could be 1 or 128. But the total
> > length of the command buffer is DWORD aligned. So we designed only a 1-byte
> > data field in struct txgbe_hic_i2c_read, to avoid redundant reading and writing
> > during the SW-FW interaction.
> >
> > For 1-byte data, wx_host_interface_command() can be used to set 'return_data'
> > to true, then page->data = buffer->data. For other cases, I think it would be more
> > convenient to read directly from the mailbox registers.
> 
> With such design you always have your return data starting at offset of
> 15, which is absolutely unaligned. And then it needs more buffer
> dancing.

OK. We could consider redesigning this buffer structure. Return data will start
at offset of 16, and reserve 4 bytes. And longer data will be directly read from
the mailbox register. Is this design more reasonable?