Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <161714738634.2168142.10860201861152789544.stgit@dwillia2-desk3.amr.corp.intel.com>
 <161714740233.2168142.11116065966198937093.stgit@dwillia2-desk3.amr.corp.intel.com>
 <20210331130953.GG1463678@nvidia.com> <CAPcyv4inZaSRk-eiyeRLfUOrwyD=YVLW6bdUVJ239X099n1S=Q@mail.gmail.com>
 <20210331161759.GL1463678@nvidia.com>
In-Reply-To: <20210331161759.GL1463678@nvidia.com>
From:   Dan Williams <dan.j.williams@intel.com>
Date:   Wed, 31 Mar 2021 09:32:28 -0700
Message-ID: <CAPcyv4ifsy+e2j6EPRgy2+cBpm_8ggvaT_GPnwEQKqto+x5YYQ@mail.gmail.com>
Subject: Re: [PATCH v3 3/4] cxl/mem: Do not rely on device_add() side effects
 for dev_set_name() failures
To:     Jason Gunthorpe <jgg@nvidia.com>
Cc:     linux-cxl@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Vishal L Verma <vishal.l.verma@intel.com>,
        "Weiny, Ira" <ira.weiny@intel.com>,
        "Schofield, Alison" <alison.schofield@intel.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Mar 31, 2021 at 9:18 AM Jason Gunthorpe <jgg@nvidia.com> wrote:
>
> On Wed, Mar 31, 2021 at 09:04:32AM -0700, Dan Williams wrote:
> > On Wed, Mar 31, 2021 at 6:10 AM Jason Gunthorpe <jgg@nvidia.com> wrote:
> > >
> > > On Tue, Mar 30, 2021 at 04:36:42PM -0700, Dan Williams wrote:
> > > > +static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
> > > > +{
> > > > +     struct cxl_memdev *cxlmd;
> > > > +     struct device *dev;
> > > > +     struct cdev *cdev;
> > > > +     int rc;
> > > > +
> > > > +     cxlmd = cxl_memdev_alloc(cxlm);
> > > > +     if (IS_ERR(cxlmd))
> > > > +             return PTR_ERR(cxlmd);
> > > > +
> > > > +     dev = &cxlmd->dev;
> > > > +     rc = dev_set_name(dev, "mem%d", cxlmd->id);
> > > > +     if (rc)
> > > > +             goto err;
> > > >
> > > > +     cdev = &cxlmd->cdev;
> > > >       cxl_memdev_activate(cxlmd, cxlm);
> > > >       rc = cdev_device_add(cdev, dev);
> > > >       if (rc)
> > > > -             goto err_add;
> > > > +             goto err;
> > >
> > > It might read nicer to have the error unwind here just call cxl_memdev_unregister()
> >
> > Perhaps, but I don't think cdev_del() and device_del() are prepared to
> > deal with an object that was not successfully added.
>
> Oh, probably not, yuk yuk yuk.
>
> Ideally cdev_device_add should not fail in a way that allows an open,
> I think that is just an artifact of it being composed of smaller
> functions..
>
> For instance if we replace the kobj_map with xarray then we can
> use xa_reserve and xa_store to avoid this condition.
>
> This actually looks like a good fit because the dev_t has pretty
> "lumpy" allocations and this isn't really performance sensitive.
>
> A clever person could then make the dev_t self allocating and solve
> another pain point with this interface. Hum..
>

...not a bad idea.

/me bookmarks this thread for future consideration.