Re: [PATCH net-next v3] ppp: enable TX scatter-gather

On Wed, Jan 28, 2026 at 5:05 PM Paolo Abeni [off-list ref] wrote:

quoted

On Tue, Jan 27, 2026 at 8:34 PM Paolo Abeni [off-list ref] wrote:

quoted

I spent a little time trying to understanding the logic here and I think
that enabling features depending on IFF_NO_QUEUE is fragile at best.

It looks like that the IFF_NO_QUEUE bit is an inconsistent state for
multilink devices using different type of channels.
Moreover the user-space could attaching a qdisc to the ppp device after
channel initialization.

Instead you could always expose the features and linearize as needed
when transmitting on !direct_xmit channel; no need to touch the
individual channel implementation, you could do such check before
calling the ops->start_xmit() calls (possibly creating a new
wrapper/helper for that).

Attaching a new qdisc won't clear the IFF_NO_QUEUE bit. (The flag
means the interface _can_ run without a qdisc).

Correct.

quoted

As for multilink devices, one is not supposed to bundle channels with
inconsistent direct_xmit (for example, mix ppp_synctty with pptp) and
expect better results. But as the driver does not reject that, I may
add a skb_linearize() to ppp_mp_explode(), or add a check for
SC_MULTILINK flag in ppp_fix_features().

What do you think?

AFAICS nothing prevent the user-space from creating multiple channels
with different type even without SC_MULTILINK, so ppp_fix_features()
should likely check the whole channel list.

Also packets could be being transmitted after channel creation and
before the features are updated, so ppp_start_xmit could observe
transient mismatching features and skb layout.

It's not a matter of bad performances: if skb is not linear and the
channel start_xmit assumes linear layout bad things will happen. I think
that even with a correct ppp_fix_features() you will need to check for
linearization in the datapath (as an unlikely condition).

Fair enough. I'll send v4 with your proposed changes.

Regards,
Qingfang