On 2023/05/03 16:46, Sriram Yagnaraman wrote:
-----Original Message-----
From: Tomasz Dzieciol/VIM Integration (NC) /SRPOL/Engineer/Samsung
Electronics <t.dziec...@partner.samsung.com>
Sent: Tuesday, 2 May 2023 16:01
To: Sriram Yagnaraman <sriram.yagnara...@est.tech>; qemu-
de...@nongnu.org; akihiko.od...@daynix.com
Cc: jasow...@redhat.com; k.kwiec...@samsung.com;
m.socha...@samsung.com
Subject: RE: [PATCH v3 1/2] igb: RX descriptors handling cleanup
Not Linux/DPDK/FreeBSD for IGB.
Change here adds additional condition (RXCSUM.IPPCSE set) to enable putting
IP ID into descriptor, besides clearing RXCSUM.PCSD (required according to
Intel 82576 datasheet) that was not present in the e1000e code.
Yes, we can't even use ethtool to set this field.
My suggestion is to not add/maintain code that we cannot test. I leave it up to
Akhikho to decide if we really need to implement IPPCSE.
The default value of RXCSUM.IPPCSE is unset, so we could as well ignore this
field until there is a user who sets this.
In general I won't reject a patch to implement a feature not used by a
known guest, but I don't recommend that. It just doesn't make sense to
spend time to write code that can turn out so buggy that it is unusable
in practice, which is often the case with untested code.
