On Thu, 08/21 14:19, Paolo Bonzini wrote: > Il 21/08/2014 13:56, Fam Zheng ha scritto: > > We are blocking the whole VM, which means that an irresponsive storage > > backend will hang the whole guest. Let's switch to bdrv_aio_cancel_async > > to improve this. > > Unforuntately, the TMF must only return after the request has been > canceled. I think you need to add a scsi_cancel_io_async function, and > keep all the remaining machinery (also, you need a better commit message > that explains what you are removing and the new invariants). > > Then in virtio-scsi you need to add a list of "dependent" (controlq) > VirtIOSCSIReq to the "main" (requestq) VirtIOSCSIReq, and complete them > all after signaling the completion of the main request.
OK, I didn't know that. I'll try again :) Thanks, Fam
