This patch-set mainly fixes teardown and error flow fixes that Jason and you have found in the recent submitted patchset. As suggested by Jason, I've split the cache removal process to cleanup and release. Although I don't expect the HW vendor drivers to use the cache after it was cleaned up, releasing it in the put function could prevent possible use-after-free errors. Saying that, we could alternatively review the vendors' usage more carefully and probably put the release function as part of the unregister device flow, but asynchronous contexts make it harder to review. The third patch in this series was written by Jason. The current implementation of ib_register_device could have caused the device to be released if an error occurs. By then, a consumer would call ib_dealloc_device and get a double-free bug. This patch makes the following schema mandatory for all flows: ib_alloc_device ib_register_device ib_unregister_device ib_dealloc_device In addition, this patch-set also fixes a small error flow issue that was found by Dan Carpenter's kbuild system and a possible dead-lock. Except for patch 0003, these patches could be squashed into "IB/core: Add RoCE GID table management". If you choose to do so, patch 0003 should come before this series. I can of course re-send the RoCE GID table management series if needed. Thanks, Matan Jason Gunthorpe (1): IB/core: Make ib_alloc_device initialize the kobject Matan Barak (3): IB/core: Access to one past end of array in _gid_table_setup_one IB/core: Fix possible deadlock in write_gid IB/core: RoCE GID management separate cleanup and release drivers/infiniband/core/cache.c | 109 +++++++++++++++++++++++++----------- drivers/infiniband/core/core_priv.h | 4 +- drivers/infiniband/core/device.c | 109 +++++++++++++++++++++++------------- drivers/infiniband/core/sysfs.c | 52 ++--------------- 4 files changed, 151 insertions(+), 123 deletions(-) -- 2.1.0 -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
