Re: Regmap and register ranges

Re: Regmap and register ranges

[Dmitry Torokhov]

On Fri, May 17, 2013 at 09:50:25PM +0100, Mark Brown wrote:
> On Fri, May 17, 2013 at 09:24:30AM -0700, Dmitry Torokhov wrote:
> 
> > I am looking at regmap and I wonder if I am missing something with the
> > way it checks if a window crosses another range boundary or if a
> > selector is in some other window as it seem to be always tripping on
> > itself. Should we skip the current range as we iterate through them
> > and do sanity checks? 
> 
> > The setup: a device with 0x0000 - 0xffff potential register range, paged
> > in chunks of 256 registers with page selector register repeated at 0xff.
> 
> The general idea is that the pages should be mapped in to virtual
> addresses rather than real ones so in your example you might add a
> prefix so all the registers are numbered 0x10000 to 0x1ffff.  The
> physical registers and the ones inside the paged range aren't supposed
> to overlap.

The issue I have with doing an arbitrary offset is that then data
exposed in debugfs does not match the data from device's specification.

> 
> We should probably either do sanity checks for this at init time or come
> up with a sensible way to support this in case someone does decide to do
> it, though this is the first time anyone noticed.

Does this mean you'll be open to patches allowing this scenario?

Thanks.

-- 
Dmitry

Re: Regmap and register ranges

[Mark Brown]

[-- Attachment #1: Type: text/plain, Size: 1166 bytes --]

On Fri, May 17, 2013 at 02:52:46PM -0700, Dmitry Torokhov wrote:
> On Fri, May 17, 2013 at 09:50:25PM +0100, Mark Brown wrote:

> > The general idea is that the pages should be mapped in to virtual
> > addresses rather than real ones so in your example you might add a
> > prefix so all the registers are numbered 0x10000 to 0x1ffff.  The
> > physical registers and the ones inside the paged range aren't supposed
> > to overlap.

> The issue I have with doing an arbitrary offset is that then data
> exposed in debugfs does not match the data from device's specification.

Adding a random prefix on the front does mean you can just ignore the
prefix.  But this seems a fairly odd way for the device to do things to
be honest...  paging every single register is a bit random.

> > We should probably either do sanity checks for this at init time or come
> > up with a sensible way to support this in case someone does decide to do
> > it, though this is the first time anyone noticed.

> Does this mean you'll be open to patches allowing this scenario?

Obviously I'm always open to patches that aren't crazy, though I do
reserve the right to define what's crazy :)

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]