On Fri, May 07, 2021 at 06:59:01PM +0200, Greg Kurz wrote:
> Now that virtio-scsi-pci and virtio-blk-pci map 1 virtqueue per vCPU,
> a serious slow down may be observed on setups with a big enough number
> of vCPUs.
> 
> Exemple with a pseries guest on a bi-POWER9 socket system (128 HW threads):
> 
>               virtio-scsi      virtio-blk
> 
> 1		0m20.922s	0m21.346s
> 2		0m21.230s	0m20.350s
> 4		0m21.761s	0m20.997s
> 8		0m22.770s	0m20.051s
> 16		0m22.038s	0m19.994s
> 32		0m22.928s	0m20.803s
> 64		0m26.583s	0m22.953s
> 128		0m41.273s	0m32.333s
> 256		2m4.727s 	1m16.924s
> 384		6m5.563s 	3m26.186s
> 
> Both perf and gprof indicate that QEMU is hogging CPUs when setting up
> the ioeventfds:
> 
>  67.88%  swapper         [kernel.kallsyms]  [k] power_pmu_enable
>   9.47%  qemu-kvm        [kernel.kallsyms]  [k] smp_call_function_single
>   8.64%  qemu-kvm        [kernel.kallsyms]  [k] power_pmu_enable
> =>2.79%  qemu-kvm        qemu-kvm           [.] memory_region_ioeventfd_before
> =>2.12%  qemu-kvm        qemu-kvm           [.] address_space_update_ioeventfds
>   0.56%  kworker/8:0-mm  [kernel.kallsyms]  [k] smp_call_function_single
> 
> address_space_update_ioeventfds() is called when committing an MR
> transaction, i.e. for each ioeventfd with the current code base,
> and it internally loops on all ioventfds:
> 
> static void address_space_update_ioeventfds(AddressSpace *as)
> {
> [...]
>     FOR_EACH_FLAT_RANGE(fr, view) {
>         for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
> 
> This means that the setup of ioeventfds for these devices has
> quadratic time complexity.
> 
> This series simply changes the device models to extend the transaction
> to all virtqueueues, like already done in the past in the generic
> code with 710fccf80d78 ("virtio: improve virtio devices initialization
> time").
> 
> Only virtio-scsi and virtio-blk are covered here, but a similar change
> might also be beneficial to other device types such as host-scsi-pci,
> vhost-user-scsi-pci and vhost-user-blk-pci.
> 
>               virtio-scsi      virtio-blk
> 
> 1		0m21.271s	0m22.076s
> 2		0m20.912s	0m19.716s
> 4		0m20.508s	0m19.310s
> 8		0m21.374s	0m20.273s
> 16		0m21.559s	0m21.374s
> 32		0m22.532s	0m21.271s
> 64		0m26.550s	0m22.007s
> 128		0m29.115s	0m27.446s
> 256		0m44.752s	0m41.004s
> 384		1m2.884s	0m58.023s
> 
> This should fix https://bugzilla.redhat.com/show_bug.cgi?id=1927108
> which reported the issue for virtio-scsi-pci.
> 
> Changes since v1:
> - Add some comments (Stefan)
> - Drop optimization on the error path in patch 2 (Stefan)
> 
> Changes since RFC:
> 
> As suggested by Stefan, splimplify the code by directly beginning and
> committing the memory transaction from the device model, without all
> the virtio specific proxying code and no changes needed in the memory
> subsystem.
> 
> Greg Kurz (4):
>   virtio-blk: Fix rollback path in virtio_blk_data_plane_start()
>   virtio-blk: Configure all host notifiers in a single MR transaction
>   virtio-scsi: Set host notifiers and callbacks separately
>   virtio-scsi: Configure all host notifiers in a single MR transaction
> 
>  hw/block/dataplane/virtio-blk.c | 45 ++++++++++++++++++++-
>  hw/scsi/virtio-scsi-dataplane.c | 72 ++++++++++++++++++++++++---------
>  2 files changed, 97 insertions(+), 20 deletions(-)
> 
> -- 
> 2.26.3
> 

Thanks, applied to my block tree:
https://gitlab.com/stefanha/qemu/commits/block

Stefan