The Copy-VMFile cmdlet on the host may fail because the guest fcopy driver state machine gets out of sync. This happens because the ->state and ->context variables are accessed by the main thread and from interrupt context. If an interrupt happens between fcopy_respond_to_host and hv_poll_channel in fcopy_write, then hv_fcopy_onchannelcallback called from that interrupt sees still state HVUTIL_USERSPACE_RECV. It updates the context, but fcopy_write will not notice that update and hv_poll_channel gets called with an empty context. As a result hv_fcopy_daemon gets no more data. After a timeout Copy-VMFile fails with timeout.
In my initial testing for a fix I put a "mb()" after the last .state change in fcopy_write. But this series implementes read/write memory barriers as needed. Let me know if this is overdoing things. Dexuan Cui (1): Drivers: hv: vmbus: fix init_vp_index() for reloading hv_netvsc Olaf Hering (4): hv: add helpers to handle hv_util device state hv: fcopy: use wrappers to propagate state hv: kvp: use wrappers to propaigate state hv: vss: use wrappers to propagate state drivers/hv/channel_mgmt.c | 17 +++++++++++++++++ drivers/hv/hv_fcopy.c | 36 ++++++++++++++++++++---------------- drivers/hv/hv_kvp.c | 39 +++++++++++++++++++++------------------ drivers/hv/hv_snapshot.c | 37 ++++++++++++++++++++----------------- drivers/hv/hyperv_vmbus.h | 14 ++++++++++++++ 5 files changed, 92 insertions(+), 51 deletions(-) -- 1.7.4.1 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
