On Fri, Sep 5, 2025 at 8:22 AM Stanislav Fomichev stfomichev@gmail.com wrote:
On 09/04, Mina Almasry wrote:
On Thu, Sep 4, 2025 at 11:27 AM Stanislav Fomichev sdf@fomichev.me wrote:
devmem test fails on NIPA. Most likely we get skb(s) with readable frags (why?)
I would expect if we get readable frags that the frags land in the host buffer we provide ncdevmem and we actually hit this error:
1 if (!is_devmem) { 0 pr_err("flow steering error"); 1 goto err_close_client; 2 }which as it says, should be root caused in a flow steering error. I don't know what would cause an EFAULT off the top of my head.
Yea, I don't understand what happens :-( I'm thinking of doing the following as well:
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 40b774b4f587..0c18a8c7965f 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -2820,7 +2820,7 @@ static int tcp_recvmsg_locked(struct sock *sk, struct msghdr *msg, size_t len, used); if (err <= 0) { if (!copied)
copied = -EFAULT;
copied = err; break; }Should give us more info for the devmem case... LMK if you don't like it. If I don't hear from you in a couple of days, I'll send it out..
Hmm, the other code paths overwrite the error to EFAULT; I don't know if that's significant in some way. But seems fine to me, I don't see why not do this, other than maybe potentional confusion with recvmsg returning an error not documented here:
https://linux.die.net/man/2/recvmsg
But that seems a minor point.