Re: [PATCH v8 02/10] nvdimm: Add wrapper for IOCTL pass thru

From: Jerry Hoemann <jerry.hoemann@hpe.com>
To: Dan Williams <dan.j.williams@intel.com>
Cc: "linux-nvdimm@lists.01.org" <linux-nvdimm@lists.01.org>
Subject: Re: [PATCH v8 02/10] nvdimm: Add wrapper for IOCTL pass thru
Date: Mon, 11 Apr 2016 18:19:37 -0600	[thread overview]
Message-ID: <20160412001937.GE119165@tevye.fc.hp.com> (raw)
In-Reply-To: <CAPcyv4g8BLp=oQhENAHeHaD0w5=vqu_xHXkXFbBE+xLJ64ohSA@mail.gmail.com>

On Mon, Apr 11, 2016 at 04:23:55PM -0700, Dan Williams wrote:
> On Mon, Apr 11, 2016 at 4:16 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote:
> > On Mon, Apr 11, 2016 at 11:21:45AM -0700, Dan Williams wrote:
> >> On Mon, Mar 21, 2016 at 12:37 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote:
> >> > +
> >> > +/*
> >> > + * the list nd_family of commands.  The mapping to firmware handles
> >> > + * is definied in the nfit_cmd_family_tbl
> >> > + *
> >> > + */
> >> > +#define ND_TYPE_BUS                    1
> >> > +#define ND_TYPE_DIMM_INTEL1            2
> >> > +#define ND_TYPE_DIMM_N_HPE1            3
> >> > +#define ND_TYPE_DIMM_N_HPE2            4
> >> > +
> >> > +
> >> > +#define ND_IOCTL_CALL                  _IOWR(ND_IOCTL, ND_CMD_CALL,\
> >> > +                                       struct nd_cmd_pkg)
> >> > +
> >> > +
> >> > +
> >> >  #endif /* __NDCTL_H__ */
> >>
> >> Looks, good, but for now lets assume there will never be such a thing
> >> as ND_TYPE_BUS commands or an ND_CMD_CALL for the bus.  I.e. all the
> >> commands that currently exist for the bus are commands that the kernel
> >> wants to handle.  In this way, struct nd_cmd_pkg is a solution to the
> >> problem of commands that the kernel knows exist, but has no use case
> >> to generate or parse itself.  Commands that fit that description
> >> currently only exist in leaf node implementations.
> >
> >
> > Would user space application ever want to call commands for the bus?
> >
> > I assume the answer is yes, as there are ioctls for ND_CMD_ARS_CAP, etc.,
> >
> > My thoughts here was to make the pass thru uniform, capable of calling
> > all the dsm functions that a user space agent might want to call.   I.e
> > don't make user call some _DSM functions one way, and other DSM function
> > another way....
> >
> 
> Indeed that would have been nice had that been the implementation from
> the beginning, but binaries using the current more verbose interface
> are already shipping in distributions.  So, the train has already left

This doesn't break existing binaries. They will still work.

I'm looking at the world going forward.  :)

> the station on having a "one ioctl interface to rule them all"
> interface.
> 
> Unless the ACPI working group adds a slew of kernel irrelevant
> commands to the bus, I don't see nd_cmd_pkg being needed at that root
> device level.

-- 

-----------------------------------------------------------------------------
Jerry Hoemann                  Software Engineer   Hewlett Packard Enterprise
-----------------------------------------------------------------------------
_______________________________________________
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm