svn commit: r234723 - in head/sys: amd64/amd64 amd64/include i386/i386 i386/include pc98/pc98 sparc64/include sparc64/sparc64 x86/x86

[John Baldwin]

On Friday, April 27, 2012 10:23:14 am Attilio Rao wrote:
> 2012/4/27, John Baldwin <jhb at freebsd.org>:
> > On Friday, April 27, 2012 9:55:41 am Attilio Rao wrote:
> >> 2012/4/27, John Baldwin <jhb at freebsd.org>:
> >> > On Thursday, April 26, 2012 4:24:25 pm Attilio Rao wrote:
> >> >> Author: attilio
> >> >> Date: Thu Apr 26 20:24:25 2012
> >> >> New Revision: 234723
> >> >> URL: http://svn.freebsd.org/changeset/base/234723
> >> >>
> >> >> Log:
> >> >>   Clean up the intr* MD KPI from the SMP dependency, removing a cause
> >> >> of
> >> >>   discrepancy between modules and kernel, but deal with SMP
> >> >> differences
> >> >>   within the functions themselves.
> >> >>
> >> >>   As an added bonus this also helps in terms of code readability.
> >> >
> >> > Hmm, this should not have affected anything with modules as no modules
> >> > should have ever called this.  Also, making intr_bind() available for
> >> > UP
> >> > kernels on x86 is largely pointless.  It's only caller is already
> >> > conditional
> >> > in sys/x86/x86/nexus.c:
> >>
> >> It doesn't mean that there couldn't be an user in the future of this
> >> KPI in terms of thirdy part modules.
> >> These functions are part of the public KPI and we speak a lot about
> >> making our KPI as less as dependent by compiling options so I don't
> >> really see why you are not happy about this.
> >>
> >> If the function (intr_bind() included) is part of the public KPI it
> >> can be called by thirdy part modules too.
> >
> > No!  intr_bind() is _NOT_ part of the public KPI.  The public KPI is
> > bus_intr_bind() which accepts a struct resource, which is the approved
> > handle for interrupt resources in drivers and other modules.
> 
> It is accessible from machine/intr_machdep.h so *it is* part of the
> public KPI by definition.

I will argue just this one point.  I don't think this is true.  There are many
things in machine/foo headers that I do not consider to be part of the public 
KPI, but are solely for use in MD code, for example for the nexus(4) driver to 
interface with other MD code such as msi.c or intr_machdep.c on x86.  Almost 
all of <machine/intr_machdep.h> falls into this category.  In fact, I don't 
see a single thing in the current <machine/intr_machdep.h> that is intended to 
be used by any MI code and is part of the public KPI.

-- 
John Baldwin