On Wed, 18 Nov 2020 23:40:40 -0600 Tom Seewald wrote:
> On Tue, Nov 17, 2020 at 4:26 PM Jakub Kicinski <k...@kernel.org> wrote:
> >
> > On Sun, 15 Nov 2020 20:31:40 -0600 Tom Seewald wrote:
> > > After commit 9d2e5e9eeb59 ("cxgb4/ch_ktls: decrypted bit is not enough")
> > > building the kernel with CHELSIO_T4=y and CHELSIO_TLS_DEVICE=n results
> > > in the following error:
> > >
> > > ld: drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.o: in function
> > > `cxgb_select_queue':
> > > cxgb4_main.c:(.text+0x2dac): undefined reference to
> > > `tls_validate_xmit_skb'
> > >
> > > This is caused by cxgb_select_queue() calling cxgb4_is_ktls_skb() without
> > > checking if CHELSIO_TLS_DEVICE=y. Fix this by calling cxgb4_is_ktls_skb()
> > > only when this config option is enabled.
> > >
> > > Fixes: 9d2e5e9eeb59 ("cxgb4/ch_ktls: decrypted bit is not enough")
> > > Signed-off-by: Tom Seewald <tseew...@gmail.com>
> > > ---
> > > drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 4 +++-
> > > 1 file changed, 3 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> > > b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> > > index 7fd264a6d085..8e8783afd6df 100644
> > > --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> > > +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
> > > @@ -1176,7 +1176,9 @@ static u16 cxgb_select_queue(struct net_device
> > > *dev, struct sk_buff *skb,
> > > txq = netdev_pick_tx(dev, skb, sb_dev);
> > > if (xfrm_offload(skb) || is_ptp_enabled(skb, dev) ||
> > > skb->encapsulation ||
> > > - cxgb4_is_ktls_skb(skb) ||
> > > +#if IS_ENABLED(CONFIG_CHELSIO_TLS_DEVICE)
> > > + cxgb4_is_ktls_skb(skb) ||
> > > +#endif /* CHELSIO_TLS_DEVICE */
> > > (proto != IPPROTO_TCP && proto != IPPROTO_UDP))
> > > txq = txq % pi->nqsets;
> > >
> >
> > The tls header already tries to solve this issue, it just does it
> > poorly. This is a better fix:
> >
> > diff --git a/include/net/tls.h b/include/net/tls.h
> > index baf1e99d8193..2ff3f4f7954a 100644
> > --- a/include/net/tls.h
> > +++ b/include/net/tls.h
> > @@ -441,11 +441,11 @@ struct sk_buff *
> > tls_validate_xmit_skb(struct sock *sk, struct net_device *dev,
> > struct sk_buff *skb);
> >
> > static inline bool tls_is_sk_tx_device_offloaded(struct sock *sk)
> > {
> > -#ifdef CONFIG_SOCK_VALIDATE_XMIT
> > +#ifdef CONFIG_TLS_DEVICE
> > return sk_fullsock(sk) &&
> > (smp_load_acquire(&sk->sk_validate_xmit_skb) ==
> > &tls_validate_xmit_skb);
> > #else
> > return false;
> >
> >
> > Please test this and submit if it indeed solves the problem.
> >
> > Thanks!
>
> Hi Jakub,
>
> Thanks for the reply, unfortunately that patch does not resolve the
> issue, I still get the same error as before. After looking into this a
> bit further, the issue seems to be with CONFIG_TLS=m as everything
> works when CONFIG_TLS=y.
>
> I also see that there was a similar issue [1] reported by Intel's
> kbuild test robot where the cxgb4 driver isn't able to see the TLS
> symbols when CONFIG_TLS=m.
Interesting. Does your original patch solve the allyesconfig + TLS=m
problem?
Seems to me that CHELSIO_T4 should depend on (TLS || TLS=n), the
CONFIG_CHELSIO_TLS_DEVICE has the dependency but AFAICT nothing prevents
CONFIG_CHELSIO_TLS_DEVICE=m and CHELSIO_T4=y and cxgb4_main.c is under
the latter.
