The value is set to uint16_t by mistake. The value is passed to Qemu in bytes, but then it is divided by the sector size and passed to the vm in sectors through a 16 bit register field. The vm kernel then multiplies it again by sector size and shows (through /sys/block/x/queue) the value in bytes. Because of this, the input value from the qemu side should be uint32_t.
A simple patch is attached, correcting the problem. I think some extra logic should be added after dividing by the page size, to prevent overflowing the 16 bit register. With that patch, I can pass a 4MB min and optimal io sizes successfully, to match the stripe size used by RBD/Ceph. Sadly though, I see no improvement coming out of it, but I'll document that in a separate bug report. root@cephvm:~# lsblk -t NAME ALIGNMENT MIN-IO OPT-IO PHY-SEC LOG-SEC ROTA SCHED RQ-SIZE RA WSAME sda 0 4194304 4194304 512 512 1 noop 128 4096 2G ** Patch added: "min_io_size32.patch" https://bugs.launchpad.net/qemu/+bug/1600563/+attachment/4915272/+files/min_io_size32.patch -- You received this bug notification because you are a member of qemu- devel-ml, which is subscribed to QEMU. https://bugs.launchpad.net/bugs/1600563 Title: min_io_size is currently limited to size uint16_t Status in QEMU: New Bug description: I am using LVM VGs on MD-raid1 for hosting my KVM volumes. On the host, a VG looks like this: Disk /dev/vm/vol202a: 60 GiB, 64424509440 bytes, 125829120 sectors Units: sectors of 1 * 512 = 512 bytes Sector size (logical/physical): 512 bytes / 4096 bytes I/O size (minimum/optimal): 131072 bytes / 131072 bytes In order to replicate the block device characteristics in the guest, I am using the following extra parameters: -set device.scsi0-0-0-0.logical_block_size=512 -set device.scsi0-0-0-0.physical_block_size=4096 -set device.scsi0-0-0-0.min_io_size=131072 This doesn't work as qemu complains that min_io_size needs to be of type uint16_t. To manage notifications about this bug go to: https://bugs.launchpad.net/qemu/+bug/1600563/+subscriptions