Re: [PATCH 1/2] UIO: add device clock support

[Paul Mundt <lethal@linux-sh.org>]

On Fri, Jun 26, 2009 at 12:17:03AM +0900, Magnus Damm wrote:
> On Thu, Jun 25, 2009 at 3:41 AM, Paul Mundt<lethal@linux-sh.org> wrote:
> > On Wed, Jun 24, 2009 at 09:34:14AM -0700, Greg KH wrote:
> >> On Wed, Jun 24, 2009 at 05:30:24PM +0200, Daniel Mack wrote:
> >> > Add a pointer to a 'struct clk' to uio_info. Drivers can set
> >> > this pointer if a clock is needed, and the UIO core will care
> >> > to enable and disable it upon device open and release.
> >>
> >> Do you have a UIO driver that needs this?
> >>
> >> If so, please submit it at the same time, otherwise adding
> >> infrastructure for no driver that needs it, is pretty pointless.
> >>
> > We can use this for the uio_pdrv_genirq case on sh, but open/release is a
> > bit coarse grained. Presently we default-enable clocks for devices that
> > are handled through uio, so doing it at open/release is at least better
> > than that. The other thing to consider is if we really want to add a
> > HAVE_CLK depdendency outright, or just ifdef around it..
> 
> Yeah, we export quite a few UIO devices on SuperH, and
> stopping/starting clocks is on my TODO list.
> 
> Like Paul says, open/release only is a bit coarse grained. I'd like to
> be able to keep the device open in user space, having a bunch of
> memory windows mmap()ed but still being able to power down the
> hardware block somehow.
> 
> The patch seems to be about enabling and disabling clocks, the actual
> clock frequency is not really exposed to user space what I can tell. I
> was more thinking along the lines of having an array of clocks per uio
> device - pretty much like the memory windows - and just expose the
> clocks and their frequencies to user space and letting user space
> enable and disable using sysfs.
> 
This is longer term work that still needs to be discussed and agreed
upon, which is totally out of scope for this patch. Doing it at
open/close time is already better than what we have today, so I am
inclined to roll this patch in to a topic branch and convert the existing
sh users over to it. Any improvement with regards to making it more fine
grained, or deciding what to expose to userspace, is something to look at
much later, particularly as the runtime pm code is going to be a
prerequisite for a lot of it (ie, 2.6.33).

Greg, assuming you have no objections, if you can ack the patch I can
carry it in my tree and do the sh conversion on top of it. It will
require HAVE_CLK ifdefs for now, but I'll take care of that.