On Mon, Mar 10, 2025 at 11:55 AM Jason Wang jasowang@redhat.com wrote:
On Fri, Mar 7, 2025 at 7:01 PM Akihiko Odaki akihiko.odaki@daynix.com wrote:
Hash reporting
Allow the guest to reuse the hash value to make receive steering consistent between the host and guest, and to save hash computation.
RSS
RSS is a receive steering algorithm that can be negotiated to use with virtio_net. Conventionally the hash calculation was done by the VMM. However, computing the hash after the queue was chosen defeats the purpose of RSS.
Another approach is to use eBPF steering program. This approach has another downside: it cannot report the calculated hash due to the restrictive nature of eBPF steering program.
Introduce the code to perform RSS to the kernel in order to overcome thse challenges. An alternative solution is to extend the eBPF steering program so that it will be able to report to the userspace, but I didn't opt for it because extending the current mechanism of eBPF steering program as is because it relies on legacy context rewriting, and introducing kfunc-based eBPF will result in non-UAPI dependency while the other relevant virtualization APIs such as KVM and vhost_net are UAPIs.
Signed-off-by: Akihiko Odaki akihiko.odaki@daynix.com Tested-by: Lei Yang leiyang@redhat.com
Documentation/networking/tuntap.rst | 7 ++ drivers/net/Kconfig | 1 + drivers/net/tap.c | 68 ++++++++++++++- drivers/net/tun.c | 98 +++++++++++++++++----- drivers/net/tun_vnet.h | 159 ++++++++++++++++++++++++++++++++++-- include/linux/if_tap.h | 2 + include/linux/skbuff.h | 3 + include/uapi/linux/if_tun.h | 75 +++++++++++++++++ net/core/skbuff.c | 4 + 9 files changed, 386 insertions(+), 31 deletions(-)
[...]
- The %TUN_VNET_HASH_REPORT flag set with this ioctl will be effective only
- after calling the %TUNSETVNETHDRSZ ioctl with a number greater than or equal
- to the size of &struct virtio_net_hdr_v1_hash.
So you had a dependency check already for vnet hdr len. I'd still suggest to split this into rss and hash as they are separated features. Then we can use separate data structure for them instead of a container struct.
Besides this, I think we still need to add new bits to TUNGETIFF to let userspace know about the new ability.
Thanks