Hi Eric, On 2022/6/3 01:19, Eric Farman wrote:
Last autumn, Jason Gunthorpe proposed some rework of vfio-ccw [1], to better fit with the new mdev API (thank you!). Part of that series was pulled for kernel 5.16 [2], but the complexities of the remaining patches got them hung up behind other work. This series attempts to dust off and complete that, with the goal of untangling the lifecycle of a s390 subchannel when bound to vfio-ccw instead of the usual io_subchannel driver. Patches 1-8 are inspired by and/or split out from that series, in order to be consumable on their own (backports, etc.). Patches 9-12 handle the goal of making the FSM complete, and synchronizing the subchannel's life with that of the mdev. (This was the goal of patch 5 of the larger series [3].) Patches 13-14 are pulled directly from the earlier series. As these patches hit some other of the consumers of vfio, those on CC who are unfamiliar with vfio-ccw probably only care about these. :) Patches 15-18 links the lifecycle of the vfio_ccw_private struct with the mdev via a vfio reference. (Patch 17 was also pulled directly from the earlier series.) In the end, the subchannel probe/remove callbacks from the css driver simply register/unregister with vfio-mdev. The communication with the subchannel is delayed until the mdev routines, which handles all the vfio-related memory and subchannel enablement. There's no longer a configuration where the mdev is closed while the subchannel remains enabled, since that's weird. @Jason: I carried the S-o-b/r-b tags on patches 13, 14, and 17, as they were cherry-picked straight from your v3. If you'd prefer your S-o-b on others, please let me know.
very nice to see it. do you have a 5.19 based branch including these changes on github? I'd like to rebase my vfio cdev patches on top of your changes. :-)
[1] https://lore.kernel.org/r/0-v3-57c1502c62fd+2190-ccw_mdev_jgg@xxxxxxxxxx/ [2] https://lore.kernel.org/r/0-v4-cea4f5bd2c00+b52-ccw_mdev_jgg@xxxxxxxxxx/ [3] https://lore.kernel.org/r/5-v3-57c1502c62fd+2190-ccw_mdev_jgg@xxxxxxxxxx/ Cc: Kirti Wankhede <kwankhede@xxxxxxxxxx> Cc: Jonathan Corbet <corbet@xxxxxxx> Cc: Zhenyu Wang <zhenyuw@xxxxxxxxxxxxxxx> Cc: Zhi Wang <zhi.a.wang@xxxxxxxxx> Cc: intel-gvt-dev@xxxxxxxxxxxxxxxxxxxxx Cc: Tony Krowiak <akrowiak@xxxxxxxxxxxxx> Cc: Jason Herne <jjherne@xxxxxxxxxxxxx> Eric Farman (14): vfio/ccw: Fix FSM state if mdev probe fails vfio/ccw: Ensure mdev->dev is cleared on mdev remove vfio/ccw: Do not change FSM state in subchannel event vfio/ccw: Remove private->mdev vfio/ccw: Pass enum to FSM event jumptable vfio/ccw: Flatten MDEV device (un)register vfio/ccw: Check that private pointer is not NULL vfio/ccw: Create an OPEN FSM Event vfio/ccw: Create a CLOSE FSM event vfio/ccw: Refactor vfio_ccw_mdev_reset vfio/ccw: Move FSM open/close to MDEV open/close vfio/ccw: Manage private with mdev vfio/ccw: Create a get_private routine vfio/ccw: Manage ccw/mdev reference counts Jason Gunthorpe (3): vfio/mdev: Consolidate all the device_api sysfs into the core code vfio/mdev: Add mdev available instance checking to the core vfio: Export vfio_device_try_get() Michael Kawano (1): vfio/ccw: Remove UUID from s390 debug log .../driver-api/vfio-mediated-device.rst | 8 +- drivers/gpu/drm/i915/gvt/kvmgt.c | 9 +- drivers/s390/cio/vfio_ccw_async.c | 1 - drivers/s390/cio/vfio_ccw_drv.c | 114 ++++++-------- drivers/s390/cio/vfio_ccw_fsm.c | 91 +++++++++-- drivers/s390/cio/vfio_ccw_ops.c | 145 ++++++------------ drivers/s390/cio/vfio_ccw_private.h | 33 +++- drivers/s390/crypto/vfio_ap_ops.c | 41 ++--- drivers/s390/crypto/vfio_ap_private.h | 2 - drivers/vfio/mdev/mdev_core.c | 13 +- drivers/vfio/mdev/mdev_private.h | 2 + drivers/vfio/mdev/mdev_sysfs.c | 64 +++++++- drivers/vfio/vfio.c | 3 +- include/linux/mdev.h | 13 +- include/linux/vfio.h | 1 + samples/vfio-mdev/mbochs.c | 9 +- samples/vfio-mdev/mdpy.c | 31 +--- samples/vfio-mdev/mtty.c | 10 +- 18 files changed, 300 insertions(+), 290 deletions(-)
-- Regards, Yi Liu
