Date: Tue, 23 Sep 2025 20:08:59 +0000 From: bugzilla-noreply@freebsd.org To: bugs@FreeBSD.org Subject: [Bug 263229] virtio vtnet driver with rxcsum errors Message-ID: <bug-263229-227-EpXuHCMGDt@https.bugs.freebsd.org/bugzilla/> In-Reply-To: <bug-263229-227@https.bugs.freebsd.org/bugzilla/>
index | next in thread | previous in thread | raw e-mail
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=263229 --- Comment #34 from Michael Tuexen <tuexen@freebsd.org> --- (In reply to mike from comment #30) OK, I now understand what was going on: * When you have enabled hardware LRO on the vtnet driver, you forward too large packets. They trigger an ICMP response (packet too big), which is ignored, because the sender is honoring already the MTU. * The sender is doing TSO, which is generating the too large packets. However, not all packets where actually TSOed. This is fixed in: https://cgit.FreeBSD.org/src/commit/?id=1bb72d3d6eee50232b6d6acbc201f9aaac93e2b5 Now the throughput will be even less in your problematic scenario. * The crucial point is that hardware LRO should be turned off by default, and software LRO should be turned on by default. This does not interfere with IP forwarding. This is something also done by other NIC drivers supporting hardware and software LRO. The patch is under review D52684. With that you setup works just fine in the default setup (as it should). * Finally there is some cleanup to actually announce receive checksum offloading for IPv6 in review D52682. Once that is reviewed and committed, I will close this PR. -- You are receiving this mail because: You are the assignee for the bug.home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-263229-227-EpXuHCMGDt>