Re: [PATCH 1/2] User access to internal clocks

[Russell King - ARM Linux]

On Mon, Feb 09, 2009 at 06:44:40PM +0100, Davide Rizzo wrote:
> >> This driver is for user level programs to interact with system clocks.
> >> It allows to read and modify rates and parents, using virtual files.
> >> It requires the implementation of 2 additional functions in the clk interface:
> >>  clk_for_each() and clk_name().
> >> Actually I implemented that functions only for Samsung S3C24xx platform.
> >
> > NAK.
> >
> >> +       name = clk_get_name(clk);
> >
> > This implies that there is a 1:1 relationship between a 'name' and a
> > struct clk.  No such thing exists (and where it does, it's being
> > eliminated in ARM because it's just plain and simple WRONG.)
>
> Not exactly, clk_get() could work in both way: a struct clk can be
> exactly identified by name and device id (like clk_get is acting now)
> or by full name, in the format clkname.device

You're not understanding the issue(s).

1. there is no 1:1 mapping between the identifiers (struct device +
   connection ID) and the struct clk.  It's actually a many-to-one
   mapping.

   That means there is _no_ name associated with a struct clk.

   Conceptually, clk_get() gives you a struct clk for the struct device
   and a connection ID.  How that mapping is achieved isn't specified
   in the API, all that's required is that such a mapping is performed.
   It is _specifically_ intended that more than one set of {device,id}
   pairs will map to the same clk.

   So, now to insist that you can go from a struct clk to some kind of
   string identifier for it is changing this - you're now requiring
   that every struct clk has a unique name.  This is not the case.
   PXA, for instance, struct clk's are now completely nameless.  They
   have no identifier.

   If you want to have a string identifier which works in every case,
   this will:

	sprintf(identifer, "%p", clk);

   Or, I guess you could force every struct clk to have a 'sysfs_name'
   field just to export them out via sysfs - and that'll be all that
   it's used for.

2. there is no generic way to walk a set of struct clk's - indeed, there
   may be no list of them (and there exists implmentations where that is
   true) and the only list which does exist is a set ID to clk mapping
   structures.

> Otherwise, what do you suggest to enumerate and distinguish all system
> clocks ?

There exists no such concept in the API, what you're asking for is
implementation specific.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/