[Qemu-devel] using MSIX for virtio-blk

[Qemu-devel] using MSIX for virtio-blk

[Naredula Janardhana Reddy]

[-- Attachment #1: Type: text/plain, Size: 178 bytes --]

Hi,
 How to enable MSIX for virtio-blk, this is to avoid extra use of
VIRTIO_PCI_ISR port reads.
This is for qemu 2.3.0 with virtio-blk-pci with x-data-plane is on.

Thanks
Jana

[-- Attachment #2: Type: text/html, Size: 220 bytes --]

Re: [Qemu-devel] using MSIX for virtio-blk

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 1235 bytes --]

On Tue, Jul 07, 2015 at 10:50:59PM +0530, Naredula Janardhana Reddy wrote:
>  How to enable MSIX for virtio-blk, this is to avoid extra use of
> VIRTIO_PCI_ISR port reads.
> This is for qemu 2.3.0 with virtio-blk-pci with x-data-plane is on.

I tried the following with QEMU 2.3.0:

  qemu-system-x86_64 -enable-kvm -m 1024 \
                     -drive if=none,id=drive0,file=test.img,format=raw,aio=native,cache=none \
		     -object iothread,id=iothread0 \
		     -device virtio-blk-pci,drive=drive0,iothread=iothread0

Inside the guest I checked /proc/interrupts and found that the
virtio-pci/virtio-blk driver is using MSI.

I also checked /proc/ioports and found the virtio-pci registers at
0xc040.

Then I used "perf record -a -e kvm:kvm_pio" on the host to record all
PIO accesses.

Inside the guest I ran "dd if=/dev/vda of=/dev/null iflag=direct bs=4k"
for about 30 seconds.  Afterwards I ran "perf script" on the host and
subtracted the 0xc040 base address from the PIO accesses to identify the
virtio-pci registers.

I didn't see any accesses to the ISR register, only writes to the
QUEUE_NOTIFY register.

Either your guest operating system is not enabling MSI or you are using
QEMU command-line options that prevent MSI.

[-- Attachment #2: Type: application/pgp-signature, Size: 473 bytes --]

Re: [Qemu-devel] using MSIX for virtio-blk

[Naredula Janardhana Reddy]

[-- Attachment #1: Type: text/plain, Size: 1512 bytes --]

Thanks Stefan,
    It's a small bug in the virtio-blk driver  in my  kernel. After fixing
, It works as expected.

Jana

On 9 July 2015 at 16:31, Stefan Hajnoczi <stefanha@gmail.com> wrote:

> On Tue, Jul 07, 2015 at 10:50:59PM +0530, Naredula Janardhana Reddy wrote:
> >  How to enable MSIX for virtio-blk, this is to avoid extra use of
> > VIRTIO_PCI_ISR port reads.
> > This is for qemu 2.3.0 with virtio-blk-pci with x-data-plane is on.
>
> I tried the following with QEMU 2.3.0:
>
>   qemu-system-x86_64 -enable-kvm -m 1024 \
>                      -drive
> if=none,id=drive0,file=test.img,format=raw,aio=native,cache=none \
>                      -object iothread,id=iothread0 \
>                      -device virtio-blk-pci,drive=drive0,iothread=iothread0
>
> Inside the guest I checked /proc/interrupts and found that the
> virtio-pci/virtio-blk driver is using MSI.
>
> I also checked /proc/ioports and found the virtio-pci registers at
> 0xc040.
>
> Then I used "perf record -a -e kvm:kvm_pio" on the host to record all
> PIO accesses.
>
> Inside the guest I ran "dd if=/dev/vda of=/dev/null iflag=direct bs=4k"
> for about 30 seconds.  Afterwards I ran "perf script" on the host and
> subtracted the 0xc040 base address from the PIO accesses to identify the
> virtio-pci registers.
>
> I didn't see any accesses to the ISR register, only writes to the
> QUEUE_NOTIFY register.
>
> Either your guest operating system is not enabling MSI or you are using
> QEMU command-line options that prevent MSI.
>

[-- Attachment #2: Type: text/html, Size: 2073 bytes --]