Re: [PATCH net v4 0/2] net: ravb: fix PTP clock lifetime

Hi Niklas,

在 2026/8/13 18:41, Niklas Söderlund 写道:

> Hi Xuanqiang,
>
> Thanks for your work.
>
> On 2026-08-11 18:37:31 +0800, Xuanqiang Luo wrote:
>> From: Xuanqiang Luo <[email protected]>
>>
>> This series fixes RAVB PTP clock lifetime handling. It reports a cached PHC
>> index without accessing the clock pointer and drains PTP interrupts before
>> unregistering the clock.
>>
>> Patch 1 caches the PHC index and handles registration failures.
>>
>> Patch 2 detaches the clock with xchg() and drains the PTP IRQs before
>> unregistering it.
> These patches are rather big change adding READ_ONCE() and WRITE_ONCE()
> to avoid a LLM warning? Or have you hit a real issue? How have you
> tested this work?

The READ_ONCE()/WRITE_ONCE() use is as follows.

PATCH 1 follows Vadim's suggestion to cache the PHC index.
ravb_get_ts_info() can race with ravb_ptp_init(), so the accessors
are there for that update.

PATCH 2 only addresses the race reported by Sashiko. The teardown
follows Vadim's suggestion as well, and READ_ONCE()/WRITE_ONCE()
looked appropriate there too.

I do not have RAVB hardware, so I have not reproduced the PATCH 2
issue.

> If you have a test-case could you share it? I have a pending series [1]
> that cleans up the whole RAVB driver ptp management which have grown
> rather organically. It have a small fix for the missing check of
> registering the clock. Would it be possible for you to test your work
> with that series too?
>
> 1.  https://lore.kernel.org/all/20260811160200.2049987-1-niklas.soderlund%[email protected]/

I looked at [1]. It is for net-next. But my goal is still a small fixfor 
net.I would prefer to keep this on net rather than rebase onto [1] for 
net-next. Does that work for you?Patch 8 of [1] moves the PHC index 
lookup into a callback, but that
callback still does not check for a NULL clock. Userspace can run
"ethtool -T" before the interface is opened, so the original hole
is still there.

To keep the conflict small, may I drop PATCH 2 for now and send
only PATCH 1 to net? PATCH 2 is a larger change, and I have no
reproducer, even though it does address Sashiko's report. The race
can wait until it is actually hit. I am not sure this is theright 
approach; this is the first time I have been in thissituation.

Thanks,
Xuanqiang