Thread (26 messages) read the whole thread 26 messages, 6 authors, 3d ago

Re: [PATCH v4 net-next 1/7] ptp: Add ioctls for PHC timestamps with quality attributes

From: Arthur Kiyanovski <akiyano@amazon.com>
Date: 2026-07-29 00:27:14
Also in: linux-doc, linux-kselftest

On 2026-07-28 11:10:31+03:00, Carolina Jubran wrote:
On 16/07/2026 21:14, Jacob Keller wrote:
quoted
On 7/16/2026 1:09 AM, Arthur Kiyanovski wrote:
Right.

Absolutely. I don't think that should change this patch series. Its just
a thought that we might want to tackle this at some point as the ioctl
interface is clearly reaching its limits.

Makes sense. Leave it up to userspace to coordinate and combine relevant
data from the device/driver and the daemons together. Ok.
That assumes host userspace knows about the synchronizer. That is not
always true.
On a DPU/SmartNIC, linuxptp may run on the device side while the host
has no visibility into that process. There is then nothing for host
userspace to coordinate with, and no way to learn the clock quality.
What should happen in that case? Would it make sense to also allow
userspace to set these attributes?
That case is actually the primary motivation for this interface, not a gap.
When the synchronizer runs device-side (DPU/SmartNIC), the device knows its
own sync quality and reports it to the host through these read-side
attributes — no host daemon to coordinate with; the host just reads
error_bound/status/timescale. ENA is exactly this: the device computes the
bound, the host reads it.

Letting userspace set these attributes is a separate feature (a host daemon
publishing quality to other host readers of the same PHC). It doesn't help
the DPU case and raises shared-clock questions (authority, permissions,
races), so I'd keep this series read-only and explore a set-side path as a
follow-up if there's demand.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help