On 8/5/2015 3:23 AM, Jason Gunthorpe wrote:
On Tue, Aug 04, 2015 at 05:03:28PM +0300, Yishai Hadas wrote:
Currently, IB/cma remove_one flow blocks until all user descriptor managed by
IB/ucma are released. This prevents hot-removal of IB devices. This patch
allows IB/cma to remove devices regardless of user space activity. Upon getting
the RDMA_CM_EVENT_DEVICE_REMOVAL event we close all the underlying HW resources
for the given ucontext. The ucontext itself is still alive till its explicit
destroying by its creator.
Implementation aside,
This changes the policy of the ucma from
Tell user space and expect it to synchronously clean up
To
Tell user space we already nuked the RDMA device asynchronously
Do we even want to do that unconditionally?
Yes, the kernel should not depend on userspace applications to approve
when it has some fatal error or device is removed/unbinded.
Shouldn't the kernel at least give userspace some time to respond to
the event before nuking the world?
No, the kernel activity is asynchronous to user-space, has higher
priority and should not wait for. The kernel raises up an event
(RDMA_CM_EVENT_DEVICE_REMOVAL) let userspace to know that the device was
removed and continue.
Application that handles events (as expected to do ..) should get it and
take relevant action items. Further application calls will result in an
error but application will stay alive.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
