When spdk_for_each_channel() was called on an io_device with no
channels, the completion function would be called immediately in the
context of the spdk_for_each_channel() caller. This is different than
the normal completion when channels are present, which is always called
asynchronously via spdk_thread_send_msg(), causing problems for callers
that did not expect to get the completion callback before
for_each_channel returned.
In particular, this causes problems with the set_bdev_qos_limit_iops RPC
method, which holds a mutex around the spdk_for_each_channel() call and
also attempts to lock that mutex again within the completion callback.
(This is GitHub issue #344.)
To avoid this problem and make the completion call consistent between
no-channels and channels-present cases, always send the completion via
spdk_thread_send_msg().
Change-Id: I9e5ec3592462c7b2ed682e665fe62ae3a5b5cc59
Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com>
Reviewed-on: https://review.gerrithub.io/417068
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
8e5eef8ebc