On Thu, Mar 31, 2016 at 8:14 AM, Lakshmi Sai Krishna Potthuri
[off-list ref] wrote:

quoted

quoted

quoted

This is really not what I'd expect to happen, I'd expect that these dummy
cycles would be in addition to the actual data (see my request for better
documentation...).  If they overlap with the data then what is the point in
specifying this?  It's more work for the host, what benefit do we get from
doing it over just handing it like a normal byte?

quoted

len field in the transfer structure contains dummy bytes along with actual

data

quoted

bytes, controllers which requires dummy bytes use len field and simply

Ignore

quoted

the dummy field (contains only no of cycles)added in this patch. Controllers
(like ZynqMP GQSPI) expects dummy in cycles won't work directly by using
len field because host driver doesn't know that len field of a particular

transfer

quoted

includes dummy bytes or not (and also number of dummy bytes included in

len

quoted

field). In such cases driver use this dummy field to identify the number of

dummy

quoted

cycles and based on that it will send the required number of dummy cycles

(which

quoted

i did in the second patch).

This doesn't make any sense at all to me.  Why does the controller care
what the bytes being sent to and from the device mean?

From the flash point of view, it expects the controller to send the dummy
on 1/2/4 lines based on the command. For Quad commands, flash expects
4 lines to be active during dummy phase. Similarly, 2 lines for dual
Commands and 1 line for normal/fast commands.
Since len field contains total number of cmd+addr+dummy bytes,
host driver should take care of sending these bytes on their respective
bus widths. Knowing when the dummy is being sent also helps in
the correct switching of IO pads (since the data lines are bidirectional)
ZynqMP GQSPI is a generic controller, majorly interfaced to flash devices.
It seems reasonable for it to know the above information from
the flash layer. Adding "dummy" cycles entry should be useful to any
controller that cares about it without affecting other spi/qspi controllers.

The m25p80 driver already uses dummy cycles, using real spi_transfer
structs, which have tx_nbits/rx_nbits fields to indicate how many data lines
to use.

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert at linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds