Re: [PATCH net 0/2] net: macb: Fix PTP timestamp configuration handling

Re: [PATCH net 0/2] net: macb: Fix PTP timestamp configuration handling

[Théo Lebrun]

Hello kimwooseok,

On Fri Sep 11, 2026 at 9:47 AM CEST, kimwooseok wrote:
> This series fixes two timestamp configuration problems in gem_set_hwtst().
> A rejected RX-filter request can change the active TX one-step mode
> while leaving the cached configuration unchanged. Patch 1 defers the
> one-step mode update until both TX type and RX filter are validated.
>
> The two specific PTPv1 RX filters currently succeed with RX timestamping
> disabled. Patch 2 routes HWTSTAMP_FILTER_PTP_V1_L4_SYNC and
> HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ through the existing all-frame
> fallback and reports HWTSTAMP_FILTER_ALL to userspace. Both defects
> originate in ab91f0a9b5f4 ("net: macb: Add hardware PTP support").
>
> Both defects were reproduced on a Raspberry Pi 5 Model B Rev 1.1 with
> macb eth0. Before the fixes, rejected requests changed OSSMODE in both
> directions; each specific PTPv1 filter produced zero hardware timestamps
> for 61 packets of its requested type. On 6.18.46-macb-ptp-functional-rt+,
> both rejected requests preserved cached settings and the full NCR, and
> each PTPv1 filter returned ALL and timestamped 60/60 Sync and 60/60
> Delay_Req fixtures. ALL/NONE controls and all nine configuration cases
> passed. The board was subsequently returned to its original kernel.
>
> Additional validation:
> - net 7f26a5e8040b: ARM64 allmodconfig and allyesconfig full targets
>   passed with GCC 14.2.0, W=1 and CONFIG_WERROR=n. After applying the
>   series, incremental builds of the same full targets passed with
>   identical configurations and no new warnings. Both rebuilt macb_ptp.o.
> - Pi board configuration: Image.gz, modules and dtbs built with W=1;
>   the resulting kernel booted and passed the hardware tests above.
> - Both final patches passed strict checkpatch including sign-off checks,
>   and standalone/series application checks on the net base.
>
> The RX tests check raw hardware timestamp presence, not absolute
> timestamp accuracy or PHC synchronization.
>
> An LLM assisted with source analysis, preparation of the fixes and test
> helpers, and drafting the descriptions. Results are from the recorded
> board tests and build logs.
>
> Assisted-by: GPT-6 Astra
>
> kimwooseok (2):
>   net: macb: Preserve one-step mode on rejected timestamp requests
>   net: macb: Use all-frame timestamping for PTPv1 RX filters
>
> base-commit: 7f26a5e8040b4957ef4dbdfcde6cc7ba2db53937

Your emails are quoted so patch emails are invalid. Please use git
send-email [0] or b4 [1] for sending patches. Both allow previewing the
messages before they are sent to the mailing lists.
For example `b4 send --no-sign --reflect` [2].

Also I much prefer cover letters and commit messages to be under the
human voice rather than their bot machine! Here at least it looks
somewhat sensible from a quick pass.

b4 even has a relay service for people with broken email servers which
might be your issue, see [3].

[0]: https://git-scm.com/docs/git-send-email
[1]: https://b4.docs.kernel.org/en/latest/
[2]: https://b4.docs.kernel.org/en/latest/contributor/send.html#checking-things-over-with-reflect
[3]: https://b4.docs.kernel.org/en/latest/contributor/send.html#configuring-the-web-endpoint

Thanks,

--
Théo Lebrun, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com