From: Cornelia Huck <cohuck@redhat.com>
To: Jason Gunthorpe <jgg@nvidia.com>, Eric Farman <farman@linux.ibm.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>,
Vasily Gorbik <gor@linux.ibm.com>,
Heiko Carstens <hca@linux.ibm.com>,
kvm@vger.kernel.org, linux-s390@vger.kernel.org,
Peter Oberparleiter <oberpar@linux.ibm.com>,
Halil Pasic <pasic@linux.ibm.com>,
Vineeth Vijayan <vneethv@linux.ibm.com>,
"Raj, Ashok" <ashok.raj@intel.com>,
Dan Williams <dan.j.williams@intel.com>,
Daniel Vetter <daniel@ffwll.ch>, Christoph Hellwig <hch@lst.de>,
Leon Romanovsky <leonro@nvidia.com>,
Max Gurtovoy <mgurtovoy@nvidia.com>,
Tarun Gupta <targupta@nvidia.com>
Subject: Re: [PATCH v2 07/13] vfio/ccw: Convert to use vfio_register_group_dev()
Date: Thu, 29 Apr 2021 13:58:55 +0200 [thread overview]
Message-ID: <20210429135855.443b7a1b.cohuck@redhat.com> (raw)
In-Reply-To: <20210428172008.GV1370958@nvidia.com>
On Wed, 28 Apr 2021 14:20:08 -0300
Jason Gunthorpe <jgg@nvidia.com> wrote:
> On Wed, Apr 28, 2021 at 07:09:49PM +0200, Cornelia Huck wrote:
> > On Mon, 26 Apr 2021 17:00:09 -0300
> > Jason Gunthorpe <jgg@nvidia.com> wrote:
> >
> > > This is more complicated because vfio_ccw is sharing the vfio_device
> > > between both the mdev_device and its vfio_device and the css_driver.
> > >
> > > The mdev is a singleton, and the reason for this sharing appears to be to
> > > allow the extra css_driver function callbacks to be delivered to the
> > > vfio_device.
> > >
> > > This keeps things as they were, with the css_driver allocating the
> > > singleton, not the mdev_driver, this is pretty confusing. I'm also
> > > uncertain how the lifetime model for the mdev works in the css_driver
> > > callbacks.
> > >
> > > At this point embed the vfio_device in the vfio_ccw_private and
> > > instantiate it as a vfio_device when the mdev probes. The drvdata of both
> > > the css_device and the mdev_device point at the private, and container_of
> > > is used to get it back from the vfio_device.
> >
> > I've been staring at this for some time, and I'm not sure whether this
> > is a good approach.
> >
> > We allow at most one mdev per subchannel (slicing it up does not make
> > sense), so we can be sure that there's a 1:1 relationship between mdev
> > and parent device, and we can track it via a single pointer.
>
> This seems like one of these cases where using the mdev GUID API was not a
> great fit. The ccs_driver should have just directly created a
> vfio_device and not gone into the mdev guid lifecycle world.
I don't remember much of the discussion back then, but I don't think
the explicit generation of devices was the part we needed, but rather
some other kind of mediation -- probably iommu related, as subchannels
don't have that concept on their own. Anyway, too late to change now.
>
> > The vfio_ccw_private driver data is allocated during probe (same as for
> > other css_drivers.) Embedding a vfio_device here means that we have a
> > structure tied into it that is operating with different lifetime rules.
> >
> > What about creating a second structure instead that can embed the
> > vfio_device, is allocated during mdev probing, and is linked up with
> > the vfio_ccw_private structure? That would follow the pattern of other
> > drivers more closely.
>
> IIRC we still end up with pointers crossing between the two
> structs. If you can't convince yourself that is correct (and I could
> not) then it is already buggy today.
>
> It is as I said to Eric, either there is no concurrency when there is
> no mdev and everything is correct today, or there is concurrency and
> it seems buggy today too.
>
> The right answer it to move the allocations out of the css_driver
> probe and put them only in the mdev driver probe because they can only
> make sense when the mdev driver is instantiated. Then everything is
> clear and very understandable how it should work.
>
> I almost did this, but couldn't figure out how the lifetime of the
> ccs_driver callbacks are working relative to the lifetime of the mdev
> device since they also reach into these structs. Maybe they can't be
> called for some css related reason?
Moving allocations to the mdev driver probe makes sense, I guess. We
should also move enabling the subchannel to that point in time (I don't
remember why we enable it in the css probe function, and can't think of
a good reason for that; obviously needs to be paired with quiescing and
disabling the subchannel in the mdev driver remove function); that
leaves the uevent dance (which can hopefully also be removed, if some
discussed changes are implemented in the common I/O layer) and fencing
QDIO.
Regarding the other callbacks,
- vfio_ccw_sch_irq should not be invoked if the subchannel is not
enabled; maybe log a message before returning for !private.
- vfio_ccw_sch_remove should be able to return 0 for !private (nothing
to quiesce, if the subchannel is not enabled).
- vfio_ccw_sch_shutdown has nothing to do for !private (same reason.)
- In vfio_ccw_sch_event, we should either skip the fsm_event and the
state change for !private, or return 0 in that case.
- vfio_ccw_chp_event already checks for !private. Not sure whether we
should try to update some control blocks and return -ENODEV if the
subchannel is not operational, but it's probably not needed.
Eric, what do you think?
next prev parent reply other threads:[~2021-04-29 11:59 UTC|newest]
Thread overview: 59+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-04-26 20:00 [PATCH v2 00/13] Remove vfio_mdev.c, mdev_parent_ops and more Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 01/13] vfio/mdev: Remove CONFIG_VFIO_MDEV_DEVICE Jason Gunthorpe
2021-04-27 11:05 ` Cornelia Huck
2021-04-26 20:00 ` [PATCH v2 02/13] vfio/mdev: Allow the mdev_parent_ops to specify the device driver to bind Jason Gunthorpe
2021-04-27 12:32 ` Cornelia Huck
2021-04-27 23:20 ` Jason Gunthorpe
2021-04-28 6:03 ` Christoph Hellwig
2021-04-28 7:56 ` Dan Williams
2021-04-28 12:41 ` Christoph Hellwig
2021-04-28 14:00 ` Jason Gunthorpe
2021-04-28 19:58 ` Dan Williams
2021-04-28 23:38 ` Jason Gunthorpe
2021-04-29 0:00 ` Dave Jiang
2021-05-26 0:42 ` Jason Gunthorpe
2021-05-26 1:42 ` Dan Williams
2021-05-27 11:44 ` Christoph Hellwig
2021-05-27 14:53 ` Jason Gunthorpe
2021-05-27 15:13 ` Christoph Hellwig
2021-04-29 6:51 ` Christoph Hellwig
2021-05-04 9:36 ` Christoph Hellwig
2021-05-04 11:30 ` Jason Gunthorpe
2021-04-28 6:44 ` Leon Romanovsky
2021-04-28 14:14 ` Jason Gunthorpe
2021-04-28 14:24 ` Leon Romanovsky
2021-04-26 20:00 ` [PATCH v2 03/13] vfio/mtty: Convert to use vfio_register_group_dev() Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 04/13] vfio/mdpy: " Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 05/13] vfio/mbochs: " Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 07/13] vfio/ccw: " Jason Gunthorpe
2021-04-27 20:06 ` Eric Farman
2021-04-27 22:10 ` Jason Gunthorpe
2021-04-28 12:55 ` Eric Farman
2021-04-28 13:21 ` Jason Gunthorpe
2021-04-28 17:09 ` Cornelia Huck
2021-04-28 17:20 ` Jason Gunthorpe
2021-04-29 11:58 ` Cornelia Huck [this message]
2021-04-29 18:13 ` Jason Gunthorpe
2021-04-30 12:31 ` Cornelia Huck
2021-04-30 17:19 ` Jason Gunthorpe
2021-05-03 10:54 ` s390 common I/O layer locking (was: [PATCH v2 07/13] vfio/ccw: Convert to use vfio_register_group_dev()) Cornelia Huck
2021-05-04 15:10 ` s390 common I/O layer locking Vineeth Vijayan
2021-07-24 13:24 ` Christoph Hellwig
2021-08-03 14:27 ` Vineeth Vijayan
2021-08-10 15:00 ` Cornelia Huck
2021-04-26 20:00 ` [PATCH v2 09/13] vfio/mdev: Remove vfio_mdev.c Jason Gunthorpe
2021-04-28 6:07 ` Christoph Hellwig
2021-04-28 6:36 ` Greg Kroah-Hartman
2021-04-28 12:53 ` Jason Gunthorpe
2021-04-29 6:53 ` Christoph Hellwig
2021-04-29 6:56 ` Greg Kroah-Hartman
2021-05-03 17:32 ` Jason Gunthorpe
2021-05-04 9:38 ` Christoph Hellwig
2021-05-04 16:20 ` Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 10/13] vfio/mdev: Remove mdev_parent_ops dev_attr_groups Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 11/13] vfio/mdev: Remove mdev_parent_ops Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 12/13] vfio/mdev: Use the driver core to create the 'remove' file Jason Gunthorpe
2021-04-26 20:00 ` [PATCH v2 13/13] vfio/mdev: Remove mdev drvdata Jason Gunthorpe
2021-04-27 21:30 ` [PATCH v2 00/13] Remove vfio_mdev.c, mdev_parent_ops and more Alex Williamson
2021-04-27 22:20 ` Jason Gunthorpe
2021-04-27 22:49 ` Alex Williamson
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210429135855.443b7a1b.cohuck@redhat.com \
--to=cohuck@redhat.com \
--cc=ashok.raj@intel.com \
--cc=borntraeger@de.ibm.com \
--cc=dan.j.williams@intel.com \
--cc=daniel@ffwll.ch \
--cc=farman@linux.ibm.com \
--cc=gor@linux.ibm.com \
--cc=hca@linux.ibm.com \
--cc=hch@lst.de \
--cc=jgg@nvidia.com \
--cc=kvm@vger.kernel.org \
--cc=leonro@nvidia.com \
--cc=linux-s390@vger.kernel.org \
--cc=mgurtovoy@nvidia.com \
--cc=oberpar@linux.ibm.com \
--cc=pasic@linux.ibm.com \
--cc=targupta@nvidia.com \
--cc=vneethv@linux.ibm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).