On Fri, Nov 04, 2016 at 10:58:04AM +0000, Stefan Hajnoczi wrote: > On Wed, Oct 19, 2016 at 04:04:50PM +0800, Fam Zheng wrote: > > The option specifies how many threads to spawn under the iothread > > object. All threads share the same AioContext so they can safely run > > (contend) together. > > > > With AioContext going away, the spawns will natually enable the block > > multi-queue work. > > > > Signed-off-by: Fam Zheng <f...@redhat.com> > > > > --- > > > > Based on v2 of Paolo's RFifoLock removal series, with which the > > symmetric contention on the single AioContext is no longer a busy > > preempt loop. > > --- > > include/sysemu/iothread.h | 19 ++++-- > > iothread.c | 148 > > ++++++++++++++++++++++++++++++++++------------ > > 2 files changed, 125 insertions(+), 42 deletions(-) > > I'm not happy with "IOThread" becoming a group of threads. IOThread > should stay the way it is. Instead you should add a new object type > that simply groups IOThreads for convenient assignment to devices, e.g. > IOThreadGroup. Then multiqueue devices can use an IOThreadGroup to work > inside multiple IOThreads.
Do we even need a IOThreadGroup object ? Can't we just explicitly pass a list of IOThread object IDs to the device. eg something like -device virtio-blk-pci,iothread=t1,iothread=t2,iothread=t3 Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://entangle-photo.org -o- http://search.cpan.org/~danberr/ :|
