Description of problem: On x86-64 (laptop, Core i7): virtio-serial upload rate: 200553666 bytes/sec (191 Mbytes/sec) virtio-serial download rate: 818819252 bytes/sec (780 Mbytes/sec) block device writes: 128171758 bytes/sec (122 Mbytes/sec) block device reads: 135930402 bytes/sec (129 Mbytes/sec) [I don't yet have a clear explanation for why guest->host (download) transfers are so much faster than the other way, but that's not important for this bug.] On aarch64: virtio-serial upload rate: 7273232 bytes/sec (6 Mbytes/sec) virtio-serial download rate: 16629760 bytes/sec (15 Mbytes/sec) block device writes: 13466009 bytes/sec (12 Mbytes/sec) block device reads: 17699635 bytes/sec (16 Mbytes/sec) Version-Release number of selected component (if applicable): x86-64 test was using qemu-2.0.0-0.1.rc0.fc21.x86_64 aarch test was with upstream qemu from git How reproducible: 100% Steps to Reproduce: 1. I'm using this program: https://github.com/libguestfs/libguestfs/blob/master/tests/qemu/qemu-speed-test.c
I suspect it's not virtio-mmio, but just KVM on ARM.
This bug appears to have been reported against 'rawhide' during the Fedora 22 development cycle. Changing version to '22'. More information and reason for this action is here: https://fedoraproject.org/wiki/Fedora_Program_Management/HouseKeeping/Fedora22
I doubt virtio-mmio is going to get any faster since all future work is looking at virtio-pci. So this bug isn't going to generate anything IMO