All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 00/10] Allow mdev drivers to directly create the vfio_device
@ 2021-06-08  0:55 ` Jason Gunthorpe
  0 siblings, 0 replies; 57+ messages in thread
From: Jason Gunthorpe @ 2021-06-08  0:55 UTC (permalink / raw)
  To: David Airlie, Tony Krowiak, Alex Williamson,
	Christian Borntraeger, Cornelia Huck, Jonathan Corbet,
	Daniel Vetter, dri-devel, Vasily Gorbik, Greg Kroah-Hartman,
	Heiko Carstens, intel-gfx, Jani Nikula, Jason Herne,
	Joonas Lahtinen, kvm, Kirti Wankhede, linux-doc, linux-s390,
	Halil Pasic, Rafael J. Wysocki, Rodrigo Vivi
  Cc: Christoph Hellwig

This is a "v3" of the previous posted full conversion:
  https://lore.kernel.org/r/0-v2-7667f42c9bad+935-vfio3_jgg@nvidia.com

Without the trailing patches that are running into complications:
 - The CCW conversion has some complicated remarks
 - AP is waiting for some locking stuff to get worked out
 - No feedback on GT
 - The license change topic for removing vfio_mdev.c

Getting the baseline functionality merged will allow Intel's IDXD mdev
driver to advance. It has already been RFC posted in the new format:

https://lore.kernel.org/kvm/162164243591.261970.3439987543338120797.stgit@djiang5-desk3.ch.intel.com/

This series includes base infrastructure and the sample conversions. The
remaining four issues can be sorted out one by one.

The major change in v3 is to enhance the driver core support for binding
based on the request from Christoph Hellwig and Dan Williams. Based on
some light analysis this looks broadly useful:

https://lore.kernel.org/kvm/20210428233856.GY1370958@nvidia.com/

====

The mdev bus's core part for managing the lifecycle of devices is mostly
as one would expect for a driver core bus subsystem.

However instead of having a normal 'struct device_driver' and binding the
actual mdev drivers through the standard driver core mechanisms it open
codes this with the struct mdev_parent_ops and provides a single driver
that shims between the VFIO core's struct vfio_device and the actual
device driver.

Instead, allow mdev drivers implement an actual struct mdev_driver and
directly call vfio_register_group_dev() in the probe() function for the
mdev. Arrange to bind the created mdev_device to the mdev_driver that is
provided by the end driver.

The actual execution flow doesn't change much, eg what was
parent_ops->create is now device_driver->probe and it is called at almost
the exact same time - except under the normal control of the driver core.

Ultimately converting all the drivers unlocks a fair number of additional
VFIO simplifications and cleanups.

v3:
 - Use device_driver_attach() from the driver core
 - 5 new patches to make device_driver_attach() exported and usable for this
 - Remove trailing patches for now
v2: https://lore.kernel.org/r/0-v2-7667f42c9bad+935-vfio3_jgg@nvidia.com
 - Keep && m in samples kconfig
 - Restore accidently squashed removeal of vfio_mdev.c
 - Remove indirections to call bus_register()/bus_unregister()
 - Reflow long doc lines
v1: https://lore.kernel.org/r/0-v1-d88406ed308e+418-vfio3_jgg@nvidia.com

Jason Gunthorpe (10):
  driver core: Do not continue searching for drivers if deferred probe
    is used
  driver core: Pull required checks into driver_probe_device()
  driver core: Flow the return code from ->probe() through to sysfs bind
  driver core: Don't return EPROBE_DEFER to userspace during sysfs bind
  driver core: Export device_driver_attach()
  vfio/mdev: Remove CONFIG_VFIO_MDEV_DEVICE
  vfio/mdev: Allow the mdev_parent_ops to specify the device driver to
    bind
  vfio/mtty: Convert to use vfio_register_group_dev()
  vfio/mdpy: Convert to use vfio_register_group_dev()
  vfio/mbochs: Convert to use vfio_register_group_dev()

 Documentation/s390/vfio-ap.rst   |   1 -
 arch/s390/Kconfig                |   2 +-
 drivers/base/base.h              |   1 -
 drivers/base/bus.c               |   6 +-
 drivers/base/dd.c                | 116 ++++++++++++-------
 drivers/gpu/drm/i915/Kconfig     |   2 +-
 drivers/vfio/mdev/Kconfig        |   7 --
 drivers/vfio/mdev/Makefile       |   3 +-
 drivers/vfio/mdev/mdev_core.c    |  46 ++++++--
 drivers/vfio/mdev/mdev_driver.c  |  10 ++
 drivers/vfio/mdev/mdev_private.h |   2 +
 drivers/vfio/mdev/vfio_mdev.c    |  24 +---
 include/linux/device.h           |   2 +
 include/linux/mdev.h             |   2 +
 samples/Kconfig                  |   6 +-
 samples/vfio-mdev/mbochs.c       | 163 +++++++++++++++------------
 samples/vfio-mdev/mdpy.c         | 159 ++++++++++++++------------
 samples/vfio-mdev/mtty.c         | 185 ++++++++++++++-----------------
 18 files changed, 397 insertions(+), 340 deletions(-)

-- 
2.31.1


^ permalink raw reply	[flat|nested] 57+ messages in thread
* Allow mdev drivers to directly create the vfio_device (v2 / alternative)
@ 2021-06-14 15:08 Christoph Hellwig
  2021-06-14 15:08 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig
  0 siblings, 1 reply; 57+ messages in thread
From: Christoph Hellwig @ 2021-06-14 15:08 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Jason Gunthorpe, Alex Williamson, Kirti Wankhede
  Cc: David Airlie, Tony Krowiak, Christian Borntraeger, Cornelia Huck,
	Jonathan Corbet, Daniel Vetter, dri-devel, Vasily Gorbik,
	Heiko Carstens, intel-gfx, Jani Nikula, Jason Herne,
	Joonas Lahtinen, kvm, linux-doc, linux-s390, Halil Pasic,
	Rafael J. Wysocki, Rodrigo Vivi

This is my alternative take on this series from Jason:

https://lore.kernel.org/dri-devel/87czsszi9i.fsf@redhat.com/T/

The mdev/vfio parts are exactly the same, but this solves the driver core
changes for the direct probing without the in/out flag that Greg hated,
which cause a little more work, but probably make the result better.

Original decription from Jason below:

The mdev bus's core part for managing the lifecycle of devices is mostly
as one would expect for a driver core bus subsystem.

However instead of having a normal 'struct device_driver' and binding the
actual mdev drivers through the standard driver core mechanisms it open
codes this with the struct mdev_parent_ops and provides a single driver
that shims between the VFIO core's struct vfio_device and the actual
device driver.

Instead, allow mdev drivers implement an actual struct mdev_driver and
directly call vfio_register_group_dev() in the probe() function for the
mdev. Arrange to bind the created mdev_device to the mdev_driver that is
provided by the end driver.

The actual execution flow doesn't change much, eg what was
parent_ops->create is now device_driver->probe and it is called at almost
the exact same time - except under the normal control of the driver core.

Ultimately converting all the drivers unlocks a fair number of additional
VFIO simplifications and cleanups.

^ permalink raw reply	[flat|nested] 57+ messages in thread
* Allow mdev drivers to directly create the vfio_device (v3)
@ 2021-06-15 13:35 Christoph Hellwig
  2021-06-15 13:35 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig
  0 siblings, 1 reply; 57+ messages in thread
From: Christoph Hellwig @ 2021-06-15 13:35 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Jason Gunthorpe, Alex Williamson, Kirti Wankhede
  Cc: David Airlie, Tony Krowiak, Christian Borntraeger, Cornelia Huck,
	Jonathan Corbet, Daniel Vetter, dri-devel, Vasily Gorbik,
	Heiko Carstens, intel-gfx, Jani Nikula, Jason Herne,
	Joonas Lahtinen, kvm, linux-doc, linux-s390, Halil Pasic,
	Rafael J. Wysocki, Rodrigo Vivi

This is my alternative take on this series from Jason:

https://lore.kernel.org/dri-devel/87czsszi9i.fsf@redhat.com/T/

The mdev/vfio parts are exactly the same, but this solves the driver core
changes for the direct probing without the in/out flag that Greg hated,
which cause a little more work, but probably make the result better.

Original decription from Jason below:

The mdev bus's core part for managing the lifecycle of devices is mostly
as one would expect for a driver core bus subsystem.

However instead of having a normal 'struct device_driver' and binding the
actual mdev drivers through the standard driver core mechanisms it open
codes this with the struct mdev_parent_ops and provides a single driver
that shims between the VFIO core's struct vfio_device and the actual
device driver.

Instead, allow mdev drivers implement an actual struct mdev_driver and
directly call vfio_register_group_dev() in the probe() function for the
mdev. Arrange to bind the created mdev_device to the mdev_driver that is
provided by the end driver.

The actual execution flow doesn't change much, eg what was
parent_ops->create is now device_driver->probe and it is called at almost
the exact same time - except under the normal control of the driver core.

Ultimately converting all the drivers unlocks a fair number of additional
VFIO simplifications and cleanups.

Changes since v1:
 - avoid warning spam for successful probes
 - improve the probe_count protection

^ permalink raw reply	[flat|nested] 57+ messages in thread
* Allow mdev drivers to directly create the vfio_device (v4)
@ 2021-06-17 14:22 Christoph Hellwig
  2021-06-17 14:22 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig
  0 siblings, 1 reply; 57+ messages in thread
From: Christoph Hellwig @ 2021-06-17 14:22 UTC (permalink / raw)
  To: Greg Kroah-Hartman, Jason Gunthorpe, Alex Williamson, Kirti Wankhede
  Cc: David Airlie, Tony Krowiak, Christian Borntraeger, Cornelia Huck,
	Jonathan Corbet, Daniel Vetter, dri-devel, Vasily Gorbik,
	Heiko Carstens, intel-gfx, Jani Nikula, Jason Herne,
	Joonas Lahtinen, kvm, linux-doc, linux-s390, Halil Pasic,
	Rafael J. Wysocki, Rodrigo Vivi

This is my alternative take on this series from Jason:

https://lore.kernel.org/dri-devel/87czsszi9i.fsf@redhat.com/T/

The mdev/vfio parts are exactly the same, but this solves the driver core
changes for the direct probing without the in/out flag that Greg hated,
which cause a little more work, but probably make the result better.

Original decription from Jason below:

The mdev bus's core part for managing the lifecycle of devices is mostly
as one would expect for a driver core bus subsystem.

However instead of having a normal 'struct device_driver' and binding the
actual mdev drivers through the standard driver core mechanisms it open
codes this with the struct mdev_parent_ops and provides a single driver
that shims between the VFIO core's struct vfio_device and the actual
device driver.

Instead, allow mdev drivers implement an actual struct mdev_driver and
directly call vfio_register_group_dev() in the probe() function for the
mdev. Arrange to bind the created mdev_device to the mdev_driver that is
provided by the end driver.

The actual execution flow doesn't change much, eg what was
parent_ops->create is now device_driver->probe and it is called at almost
the exact same time - except under the normal control of the driver core.

Ultimately converting all the drivers unlocks a fair number of additional
VFIO simplifications and cleanups.

Changes since v3:
 - minor cleanup to avoid the probe_ret variable in really_probe entirely
 - use a saner name for a variable in mdev-mtty
 - use a better driver name in mdev-mtty
 - update the documentation to match the new interface

Changes since v2:
 - avoid warning spam for successful probes
 - improve the probe_count protection

^ permalink raw reply	[flat|nested] 57+ messages in thread

end of thread, other threads:[~2021-06-17 14:24 UTC | newest]

Thread overview: 57+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-06-08  0:55 [PATCH 00/10] Allow mdev drivers to directly create the vfio_device Jason Gunthorpe
2021-06-08  0:55 ` [Intel-gfx] " Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 01/10] driver core: Do not continue searching for drivers if deferred probe is used Jason Gunthorpe
2021-06-08  5:51   ` Christoph Hellwig
2021-06-08  6:44   ` Greg Kroah-Hartman
2021-06-08 12:16     ` Jason Gunthorpe
2021-06-08 13:13       ` Greg Kroah-Hartman
2021-06-08 13:53         ` Jason Gunthorpe
2021-06-08  7:35   ` Greg Kroah-Hartman
2021-06-08 12:17     ` Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 02/10] driver core: Pull required checks into driver_probe_device() Jason Gunthorpe
2021-06-08  5:59   ` Christoph Hellwig
2021-06-08 12:21     ` Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 03/10] driver core: Flow the return code from ->probe() through to sysfs bind Jason Gunthorpe
2021-06-08  6:07   ` Christoph Hellwig
2021-06-08 23:53     ` Jason Gunthorpe
2021-06-08  6:47   ` Greg Kroah-Hartman
2021-06-08 12:30     ` Jason Gunthorpe
2021-06-08 13:16       ` Greg Kroah-Hartman
2021-06-08 14:03         ` Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during " Jason Gunthorpe
2021-06-08  6:14   ` Christoph Hellwig
2021-06-08  7:37   ` Greg Kroah-Hartman
2021-06-08  0:55 ` [PATCH 05/10] driver core: Export device_driver_attach() Jason Gunthorpe
2021-06-08  6:19   ` Christoph Hellwig
2021-06-08 12:33     ` Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 06/10] vfio/mdev: Remove CONFIG_VFIO_MDEV_DEVICE Jason Gunthorpe
2021-06-08  0:55   ` [Intel-gfx] " Jason Gunthorpe
2021-06-08  6:20   ` Christoph Hellwig
2021-06-08  6:20     ` [Intel-gfx] " Christoph Hellwig
2021-06-11 12:40   ` Cornelia Huck
2021-06-11 12:40     ` [Intel-gfx] " Cornelia Huck
2021-06-14 12:35     ` Jason Gunthorpe
2021-06-14 12:35       ` [Intel-gfx] " Jason Gunthorpe
2021-06-14 12:35       ` Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 07/10] vfio/mdev: Allow the mdev_parent_ops to specify the device driver to bind Jason Gunthorpe
2021-06-08  6:21   ` Christoph Hellwig
2021-06-08  0:55 ` [PATCH 08/10] vfio/mtty: Convert to use vfio_register_group_dev() Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 09/10] vfio/mdpy: " Jason Gunthorpe
2021-06-08  0:55 ` [PATCH 10/10] vfio/mbochs: " Jason Gunthorpe
2021-06-08  6:22 ` [PATCH 00/10] Allow mdev drivers to directly create the vfio_device Christoph Hellwig
2021-06-08  6:22   ` [Intel-gfx] " Christoph Hellwig
2021-06-14 14:34 ` Kirti Wankhede
2021-06-14 14:34   ` [Intel-gfx] " Kirti Wankhede
2021-06-14 14:34   ` Kirti Wankhede
2021-06-14 14:36   ` Jason Gunthorpe
2021-06-14 14:36     ` [Intel-gfx] " Jason Gunthorpe
2021-06-14 14:36     ` Jason Gunthorpe
2021-06-14 15:08 Allow mdev drivers to directly create the vfio_device (v2 / alternative) Christoph Hellwig
2021-06-14 15:08 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig
2021-06-14 22:43   ` Jason Gunthorpe
2021-06-14 22:43     ` Jason Gunthorpe
2021-06-15 13:35 Allow mdev drivers to directly create the vfio_device (v3) Christoph Hellwig
2021-06-15 13:35 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig
2021-06-15 14:03   ` Cornelia Huck
2021-06-15 14:03     ` Cornelia Huck
2021-06-15 19:36   ` Alex Williamson
2021-06-15 19:36     ` Alex Williamson
2021-06-17 14:22 Allow mdev drivers to directly create the vfio_device (v4) Christoph Hellwig
2021-06-17 14:22 ` [PATCH 04/10] driver core: Don't return EPROBE_DEFER to userspace during sysfs bind Christoph Hellwig

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.