This second revision takes Jens comments into account, main change is fixing an off-by-one error in patch 2.
The series introduce support for a new protocol request that notifies the backend with Virtio device status updates. Main goal is to be able with Virtio 1.0 devices to start the port even if the guest hasn't initialized all the queue pairs of the device. This case happens for example with Windows driver if more queue pairs are declared than there are vCPUs. The patch also handles reset and failed driver status to stop the device and destroy the virtqueues. Last patch implements a workaround for old and current QEMUs, that sends SET_VRING_ADDR requests for uninitalized queues, which can leads to guest memory corruption if the host application requests to diasble queues notifications. I posted the series as RFC, as the QEMU & specification parts for the new request haven't been accepted yet. Changes since RFC v1: ===================== - move virtio_status declaration in the right patch - Fix off-by-one error when removing uninitialized queues Maxime Coquelin (3): vhost: invalidate vring addresses in cleanup_vq() vhost: add SET_VIRTIO_STATUS support vhost_user: work around invalid rings addresses sent by QEMU lib/librte_vhost/vhost.c | 6 ++- lib/librte_vhost/vhost.h | 4 +- lib/librte_vhost/vhost_user.c | 113 +++++++++++++++++++++++++++++++++++++++++- lib/librte_vhost/vhost_user.h | 5 +- 4 files changed, 123 insertions(+), 5 deletions(-) -- 2.14.3
