xprtrdma: Delay DMA mapping Send and Receive buffers
authorChuck Lever <chuck.lever@oracle.com>
Thu, 15 Sep 2016 14:56:18 +0000 (10:56 -0400)
committerAnna Schumaker <Anna.Schumaker@Netapp.com>
Mon, 19 Sep 2016 17:08:37 +0000 (13:08 -0400)
commit54cbd6b0c6b9410782da3efe7377d43bb636faaf
tree15897f279f33dff79eadb4bf027c06ded70d5cd9
parent99ef4db329f1ee2413dad49346e72a6c902474d1
xprtrdma: Delay DMA mapping Send and Receive buffers

Currently, each regbuf is allocated and DMA mapped at the same time.
This is done during transport creation.

When a device driver is unloaded, every DMA-mapped buffer in use by
a transport has to be unmapped, and then remapped to the new
device if the driver is loaded again. Remapping will have to be done
_after_ the connect worker has set up the new device.

But there's an ordering problem:

call_allocate, which invokes xprt_rdma_allocate which calls
rpcrdma_alloc_regbuf to allocate Send buffers, happens _before_
the connect worker can run to set up the new device.

Instead, at transport creation, allocate each buffer, but leave it
unmapped. Once the RPC carries these buffers into ->send_request, by
which time a transport connection should have been established,
check to see that the RPC's buffers have been DMA mapped. If not,
map them there.

When device driver unplug support is added, it will simply unmap all
the transport's regbufs, but it doesn't have to deallocate the
underlying memory.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
net/sunrpc/xprtrdma/backchannel.c
net/sunrpc/xprtrdma/rpc_rdma.c
net/sunrpc/xprtrdma/verbs.c
net/sunrpc/xprtrdma/xprt_rdma.h