Re: [PATCH v4 1/6] scsi: ufs: wrap the i/o access operations

[James Bottomley <James.Bottomley@HansenPartnership.com>]

On Thu, 2013-05-09 at 12:22 +0530, Santosh Y wrote:
> On Thu, May 9, 2013 at 12:09 PM, Seungwon Jeon <tgih.jun@samsung.com> wrote:
> > On Thursday, May 09, 2013 Santosh wrote:
> >> > There are two patches remained. These are applied with your final comments.
> >> > Do you have any idea?
> >> > [PATCH v4 5/6] scsi: ufs: add dme configuration primitives
> >> > [PATCH v4 6/6] scsi: ufs: add dme control primitives
> >> >
> >>
> >> Since there is no use case for these implementations yet, except for
> >> ufshcd_get_dme_attr_val(), as per James's suggestion
> >> [http://www.mail-archive.com/linux-scsi@vger.kernel.org/msg20207.html]
> >> I did not ACK the patches.
> >> The same patches can be used to implement related features and resubmit later.
> > I respect your decision. But I have another opinion.
> > The remained patches are basic operations which should be supported by ufshcd.
> > Especially, dme_set/set will be used in vender specific part of host controller rather than in ufshcd itself.
> > And above all, Maya Erez completed to test and reported working fine.
> > If these patches are merged this time, it would be helpful to various hosts.
> >
> 
> I'm ok with merging these patches if James is fine with it. I will ACK
> the patches.

Well, no, not really.  The rule is simple: we don't add new functions to
the kernel without callers.  The reason is also simple: trying to do
interface first and then user some time later is a "make work"
development strategy that practically guarantees the interface is either
never used or needs modification.  From the kernel's point of view,
which is more important, review of a function with no callers is only
partial because you've no idea how it will be used.  Whereas if you
review a function and its callers, you can see how the API works and
possibly suggest improvements.

James