On Mon, Oct 28, 2024 at 07:42:54PM +0100, Stefano Brivio wrote:On Mon, 28 Oct 2024 20:40:44 +1100 David Gibson <david(a)gibson.dropbear.id.au> wrote:Hrm. I'm aware it theoretically need not be aligned, but I thought it was in practice.. and that we were already relying on that. In fact, I'm pretty sure the second part is true, although more subtly than here. v8 of the vhost-user patches calls tcp_fill_headers[46]() with the bp parameter set to the offset of the TCP header. If creating a tcphdr * there is a problem, then creating a tcp_payload_t * can't be any better.Currently these expects both the TCP header and payload in a single IOV, and goes to some trouble to locate the checksum field within it. In the current caller we've already know where the TCP header is, so we might as well just pass it in. This will need to work a bit differently for vhost-user, but that code already needs to locate the TCP header for other reasons, so again we can just pass it in.We couldn't do this, and also what you're now doing in 5/7, because with vhost-user the TCP header is not aligned, so we can't pass it around as a pointer, see: <ZeUpxEY-sn64NLE5@zatzit> https://archives.passt.top/passt-dev/ZeUpxEY-sn64NLE5@zatzit/ and following. That one is about IP headers, but the same applies to TCP and UDP headers.Of course the current solution is not elegant and it would be nice to find another way to deal with it, but we couldn't come up with anything better back then. The rest of the series looks good to me, but I'm afraid that without this one and 5/7 the other changes will be a bit more complicated to implement (if at all possible).Definitely. I have so ideas for approaches more robust to misalignment, but they're substantially more complicated. I was hoping we could avoid it at least for now. -- David Gibson (he or they) | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you, not the other way | around. http://www.ozlabs.org/~dgibson