Re: [GIT PULL] virtio: features

[Michael Roth]

On Sun, Sep 03, 2023 at 06:13:38PM -0400, Michael S. Tsirkin wrote:
> The following changes since commit 2dde18cd1d8fac735875f2e4987f11817cc0bc2c:
> 
>   Linux 6.5 (2023-08-27 14:49:51 -0700)
> 
> are available in the Git repository at:
> 
>   https://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git tags/for_linus
> 
> for you to fetch changes up to 1acfe2c1225899eab5ab724c91b7e1eb2881b9ab:
> 
>   virtio_ring: fix avail_wrap_counter in virtqueue_add_packed (2023-09-03 18:10:24 -0400)
> 
> ----------------------------------------------------------------
> virtio: features
> 
> a small pull request this time around, mostly because the
> vduse network got postponed to next relase so we can be sure
> we got the security store right.
> 
> Signed-off-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
> 
> ----------------------------------------------------------------
> Eugenio Pérez (4):
>       vdpa: add VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK flag
>       vdpa: accept VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK backend feature
>       vdpa: add get_backend_features vdpa operation
>       vdpa_sim: offer VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK
> 
> Jason Wang (1):
>       virtio_vdpa: build affinity masks conditionally
> 
> Xuan Zhuo (12):
>       virtio_ring: check use_dma_api before unmap desc for indirect
>       virtio_ring: put mapping error check in vring_map_one_sg
>       virtio_ring: introduce virtqueue_set_dma_premapped()
>       virtio_ring: support add premapped buf
>       virtio_ring: introduce virtqueue_dma_dev()
>       virtio_ring: skip unmap for premapped
>       virtio_ring: correct the expression of the description of virtqueue_resize()
>       virtio_ring: separate the logic of reset/enable from virtqueue_resize
>       virtio_ring: introduce virtqueue_reset()
>       virtio_ring: introduce dma map api for virtqueue
>       virtio_ring: introduce dma sync api for virtqueue
>       virtio_net: merge dma operations when filling mergeable buffers

This ^ patch (upstream commit 295525e29a) seems to cause a
network-related regression when using SWIOTLB in the guest. I noticed
this initially testing SEV guests, which use SWIOTLB by default, but
it can also be seen with normal guests when forcing SWIOTLB via
swiotlb=force kernel cmdline option. I see it with both 6.6-rc1 and
6.6-rc2 (haven't tried rc3 yet, but don't see any related changes
there), and reverting 714073495f seems to avoid the issue.

Steps to reproduce:

1) Boot QEMU/KVM guest with 6.6-rc2 with swiotlb=force via something like the following cmdline:

   qemu-system-x86_64 \
   -machine q35 -smp 4,maxcpus=255 -cpu EPYC-Milan-v2 \
   -enable-kvm -m 16G,slots=5,maxmem=256G -vga none \
   -device virtio-scsi-pci,id=scsi0,disable-legacy=on,iommu_platform=true \
   -drive file=/home/mroth/storage/ubuntu-18.04-seves2.qcow2,if=none,id=drive0,snapshot=off \
   -device scsi-hd,id=hd0,drive=drive0,bus=scsi0.0 \
   -device virtio-net-pci,netdev=netdev0,id=net0,disable-legacy=on,iommu_platform=true,romfile= \
   -netdev tap,script=/home/mroth/qemu-ifup,id=netdev0 \
   -L /home/mroth/storage/AMDSEV2/snp-release-2023-09-23/usr/local/share/qemu \
   -drive if=pflash,format=raw,unit=0,file=/home/mroth/storage/AMDSEV2/snp-release-2023-09-23/usr/local/share/qemu/OVMF_CODE.fd,readonly \
   -drive if=pflash,format=raw,unit=1,file=/home/mroth/storage/AMDSEV2/snp-release-2023-09-23/usr/local/share/qemu/OVMF_VARS.fd \
   -debugcon file:debug.log -global isa-debugcon.iobase=0x402 -msg timestamp=on \
   -kernel /boot/vmlinuz-6.6.0-rc2-vanilla0+ \
   -initrd /boot/initrd.img-6.6.0-rc2-vanilla0+ \
   -append "root=UUID=d72a6d1c-06cf-4b79-af43-f1bac4f620f9 ro console=ttyS0,115200n8 earlyprintk=serial,ttyS0,115200 debug=1 sev=debug page_poison=0 spec_rstack_overflow=off swiotlb=force"

2) scp a small file from the host to the guest IP via its virtio-net device.
   Smaller file sizes succeed, but the larger the file the more likely
   it will fail. e.g.:

   mroth@host:~$ dd if=/dev/zero of=test bs=1K count=19
   19+0 records in
   19+0 records out
   19456 bytes (19 kB, 19 KiB) copied, 0.000940134 s, 20.7 MB/s
   mroth@host:~$ scp test vm0:
   test                                                                    100%   19KB  10.1MB/s   00:00
   mroth@host:~$ dd if=/dev/zero of=test bs=1K count=20
   20+0 records in
   20+0 records out
   20480 bytes (20 kB, 20 KiB) copied, 0.00093774 s, 21.8 MB/s
   mroth@host:~$ scp test vm0:
   test                                                                      0%    0     0.0KB/s   --:-- ETA
   client_loop: send disconnect: Broken pipe
   lost connection
   mroth@host:~$

Thanks,

Mike

> 
> Yuan Yao (1):
>       virtio_ring: fix avail_wrap_counter in virtqueue_add_packed
> 
> Yue Haibing (1):
>       vdpa/mlx5: Remove unused function declarations
> 
>  drivers/net/virtio_net.c           | 230 ++++++++++++++++++---
>  drivers/vdpa/mlx5/core/mlx5_vdpa.h |   3 -
>  drivers/vdpa/vdpa_sim/vdpa_sim.c   |   8 +
>  drivers/vhost/vdpa.c               |  15 +-
>  drivers/virtio/virtio_ring.c       | 412 ++++++++++++++++++++++++++++++++-----
>  drivers/virtio/virtio_vdpa.c       |  17 +-
>  include/linux/vdpa.h               |   4 +
>  include/linux/virtio.h             |  22 ++
>  include/uapi/linux/vhost_types.h   |   4 +
>  9 files changed, 625 insertions(+), 90 deletions(-)
>