Date: Tue, 29 Aug 2017 19:09:05 +0900
From: Yasunori Goto <y-goto@jp.fujitsu.com>
To: Dan Williams <dan.j.williams@intel.com>
Subject: Re: [PATCH] libnvdimm: clean up command definitions
Cc: Jerry Hoemann <Jerry.Hoemann@hpe.com>,
        "linux-nvdimm@lists.01.org" <linux-nvdimm@lists.01.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
In-Reply-To: <CAPcyv4jp5Q9XZ=ZEp=NvUgJyh+zYoDTOGy=+4ETLjX2RDvAg-w@mail.gmail.com>
References: <20170829100344.C729.E1E9C6FF@jp.fujitsu.com> <CAPcyv4jp5Q9XZ=ZEp=NvUgJyh+zYoDTOGy=+4ETLjX2RDvAg-w@mail.gmail.com>
Message-Id: <20170829190859.F14D.E1E9C6FF@jp.fujitsu.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="US-ASCII"
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3708
Lines: 91

> On Mon, Aug 28, 2017 at 6:03 PM, Yasunori Goto <y-goto@jp.fujitsu.com> wrote:
> >> On Mon, Aug 28, 2017 at 1:50 PM, Jerry Hoemann <jerry.hoemann@hpe.com> wrote:
> >> >
> >> > On Mon, Aug 28, 2017 at 08:45:32AM -0700, Dan Williams wrote:
> >> >> Remove the command payloads that do not have an associated libnvdimm
> >> >> ioctl. I.e. remove the payloads that would only ever be carried in the
> >> >> ND_CMD_CALL envelope. This prevents userspace from growing unnecessary
> >> >> dependencies on this kernel header when userspace already has everything
> >> >> it needs to craft and send these commands.
> >> >
> >> > Userspace needs to include linux/ndctl.h to make the call as
> >> > that is where nd_cmd_pkg is defined.
> >> >
> >> > So you want to have some structures defined in ndctl.h and other
> >> > defined in the to be created libndctl-nfit.h?  Plus a third header
> >> > file for the HPE non-root calls?
> >>
> >> Yes. ndctl.h exports the ioctl command payloads, everything that goes
> >> inside of ND_CMD_CALL is defined by userspace headers. The
> >> libndctl-nfit.h header is proposed as a place to land vendor agnostic
> >> NFIT-defined payloads, and any vendor specific definitions would
> >> remain internal to libndctl as they are today.
> >>
> >> > Will libndctl-nfit.h be generally available and installed?
> >>
> >> Yes, that's the plan.
> >>
> >> > Will it be clean so that other applications can use it to get these
> >> > definitions?  Or will it be loaded w/ a bunch of stuff only useful
> >> > to your ndctl command?
> >>
> >> Yes, that's the plan. It's a bug if libndctl-nfit.h is not generically
> >> clean for issuing the NFIT root device commands via some ND_CMD_CALL
> >> helpers from the base libndctl library.
> >>
> >> In other words libndctl-nfit.h defines the payload and libndctl
> >> defines some general helpers for issuing commands.
> >
> > Maybe I don't understand your idea yet, let me confirm it.
> >
> > Certainly, current acpi driver does not need these definitions.
> > But, I think nfit_test.ko will need them to emulate these features.
> >
> > Do you intend that libndctl-nfit.h should be defined at "include/uapi/linux/"
> > directory?
> > Otherwise, it should be defined at "tools/testing/nvdimm/" or
> > "tools/testing/nvdimm/test" ?
> 
> nfit_test will need its own internal / private copy of these payloads
> in tools/testing/nvdimm/test so it can emulate how the bios behaves.
> The include/uapi/linux directory is for user to kernel interface
> definitions and these command payloads are purely an interface to bios
> / firmware.
> 

Hmm,

I'm trying to make libndctl-nfit.h under tools/testing/nvdimm/nfit/,
and trying to move NFIT_CMD_TRANSLATE_SPA from drivers/acpi/nfit/core.c to it.

But I noticed that drivers/acpi/nfit/core.c still uses the definition
in acpi_nfit_init_dsms().

----
static void acpi_nfit_init_dsms(struct acpi_nfit_desc *acpi_desc)
{
        struct nvdimm_bus_descriptor *nd_desc = &acpi_desc->nd_desc;
        const guid_t *guid = to_nfit_uuid(NFIT_DEV_BUS);

        :

        dsm_mask =
                (1 << ND_CMD_ARS_CAP) |
                (1 << ND_CMD_ARS_START) |
                (1 << ND_CMD_ARS_STATUS) |
                (1 << ND_CMD_CLEAR_ERROR) |
                (1 << NFIT_CMD_TRANSLATE_SPA) |
                (1 << NFIT_CMD_ARS_INJECT_SET) |
                (1 << NFIT_CMD_ARS_INJECT_CLEAR) |
                (1 << NFIT_CMD_ARS_INJECT_GET);
        for_each_set_bit(i, &dsm_mask, BITS_PER_LONG)
                if (acpi_check_dsm(adev->handle, guid, 1, 1ULL << i))
                        set_bit(i, &nd_desc->bus_dsm_mask);
---

Though it seems to be necessary yet,
do you have any idea to remove it?

Bye,