On 2025/2/15 0:35, Stephen Hemminger wrote:
On Fri, 14 Feb 2025 09:56:37 +0800
Jie Hai <haij...@huawei.com> wrote:
Add packet types parse for vxlan/vxlan-gpe/gtp/geneve packets.
Signed-off-by: Jie Hai <haij...@huawei.com>
---
Not sure about this.
The original purpose of ptype was to support hardware offload information.
Does any driver do this ptype detection in tunnels?
Would the software detection produce the same result as HW offload?
.
As far as I know, there are two types of packet type: sw packet type and
hw packet type.
The formmer comes from the call of function rte_net_get_ptype(), which
parses the
network headers in mbuf data and return its packet type.
All caller(of different drivers) share the same standard.
And it's commonly used in Tx checksum process.
The latter comes from the 'packet_type' field in mbuf.
It is about data really present in the Rx mbuf, and it is defined by drivers
themselves and generally arived from Rx descriptor (or hardware).
Only a few drivers use rte_net_get_ptype() to set packet_type, e.g.
1.With mask RTE_PTYPE_L2_MASK | RTE_PTYPE_L3_MASK | RTE_PTYPE_L4_MASK
drivers\net\netvsc\hn_rxtx.c
drivers\net\pfe\pfe_hif_lib.c under contidion RTE_LIBRTE_PFE_SW_PARSE
2. With mask RTE_PTYPE_ALL_MASK
drivers\net\tap\rte_eth_tap.c
drivers\net\virtio\virtio_rxtx_packed.h
drivers\net\virtio\virtio_rxtx.c
lib\vhost\virtio_net.c
lib\node\kernel_rx.c
What this patch changes is the sw packet type, only these examples are
affected.
And they may not have hardware information of packets to report.
Please check whether there is any impact, @all maintainers.
