On Tuesday 04 February 2014 03:17 PM, Thomas Gleixner wrote:

On Tue, 4 Feb 2014, Ivan Khoronzhuk wrote:

Please do not top post.

quoted

It was so in v1. But it was decided to use explicit memory barriers,
because we're always sure the memory barriers are there and that
they're properly documented. Also in this case I don't need to add
keystone readl/writel relaxed function variants and to use mixed calls of
writel/writel_relaxed functions.

See:
http://www.spinics.net/lists/arm-kernel/msg294941.html

Fair enough, but we want a proper explanation for explicit barriers in
the code and not in some random discussion of patch version X on some
random mailing list.

Aside of that it should be iowmb(), but I might miss something ...

Agree. __iowmb() seems to be more appropriate.

Regards,
Santosh